|Home | About | Journals | Submit | Contact Us | Français|
Measures of symbolic racism (SR) have often been used to tap racial prejudice toward Blacks. However, given the wording of questions used for this purpose, some of the apparent effects on attitudes toward policies to help Blacks may instead be due to political conservatism, attitudes toward government, and/or attitudes toward redistributive government policies in general. Using data from national probability sample surveys and an experiment, we explored whether SR has effects even when controlling for these potential confounds and whether its effects are specific to policies involving Blacks. Holding constant conservatism and attitudes toward limited government, SR predicted Whites' opposition to policies designed to help Blacks and more weakly predicted attitudes toward social programs whose beneficiaries were racially ambiguous. An experimental manipulation of policy beneficiaries revealed that SR predicted policy attitudes when Blacks were the beneficiary but not when women were. These findings are consistent with the claim that SR's association with racial policy preferences is not due to these confounds.
One of the most contested issues in studies of race in America concerns the role of government in addressing social inequality. To state the question at issue in its simplest form, “Exactly why do some White Americans oppose government policies designed to aid Black Americans?” Researchers have debated to what extent it is due to racial prejudice or adherence to less racial value systems (e.g., equality, individualism, and/or opposition to big government). Symbolic racism (SR) theory (McConahay & Hough, 1976; Sears, 1988; Sears & Henry, 2005; Sears & Kinder, 1971; Sears, van Laar, Carrillo, & Kosterman, 1997)—and its close relatives, such as modern racism (McConahay, 1986) and racial resentment (Kinder & Sanders, 1996)—is a particularly visible theory of racial prejudice. Although measures of SR have been used extensively throughout the social scientific literature, two particular concerns about them have been raised that merit investigation. First, we test whether the confounding of SR with race-neutral beliefs and ideologies is in fact responsible for SR's apparent effects instead of racial prejudice. We also explore whether the apparent effects of Black-focused SR on policy preferences are specific to policies that benefit Blacks in particular.
On the one hand, most White Americans appear to be dedicated to ideals of equality and true integration. But at the same time, many of those same White Americans were unwilling to support the implementation of specific policies designed to ameliorate unfair group differences. How, then, to explain this “principle-implementation gap” (Bobo, 1997; Schuman, Steeh, Bobo, & Krysan, 1997)? The tendency to oppose racial policies might still be fueled by adherence to old-fashioned, racist notions of group inferiority/superiority. However, few Whites in the past 30 years have actually subscribed to such beliefs (Bobo & Smith, 1998; Schuman et al., 1997; Sears, 1988). SR has been conceptualized as a blend of basic anti-Black antipathy and the sense that African Americans are violating consensually held values, such as working hard (Sears, 1988; Sears & Henry, 2003). SR is said to share an affective component (i.e., anti-Black affect) with old-fashioned racism. But according to the theory, Whites do not oppose racial policies because they think Blacks are subhuman or even because they are not wedded to egalitarian ideals, as was the case with old-fashioned racism (Hughes, 1997; Sears, 1988; also see Bobo & Kluegel, 1997). Rather, the SR construct has a unique component: resentment toward Blacks for not pulling themselves up out of poverty by their collective bootstraps. In other words, symbolic racism may produce opposition to such policies because (a) it contains generally negative affect toward Blacks along with (b) the belief that Blacks are not doing what all Americans should do to get ahead and that is to work hard rather than relying on the largesse of other groups or government (Sears & Henry, 2003).
Various survey questions have been devised over the years to measure SR (Henry & Sears, 2002; Kinder & Sanders, 1996; McConahay, 1986; Sears 1988). And although many would argue with the assertion that old-fashioned forms of racial prejudice had faded away (e.g., Federico & Sidanius, 2002; Sidanius, Devereux, & Pratto, 1992; Sidanius, Pratto, & Bobo, 1996), in the contemporary era the SR scales have usually predicted Whites' opposition to government policies designed to help Blacks and opposition to political candidates from ethnic minority groups better than does old-fashioned racism (see, e.g., Hughes, 1997; Kinder & Sanders, 1996; Sears, 1988; Sears et al., 1997).
The relations of SR with these outcome variables may be due to the processes outlined by SR theory, but these relations may be attributable to different processes instead (see, e.g., Carmines & Merriman, 1993; Feldman & Huddy, 2005; Roth, 1990; Sniderman & Carmines, 1997; Sniderman, Crosby, & Howell, 2000; Sniderman & Piazza, 1993; Sniderman, Piazza, Tetlock, & Kendrick, 1991; Sniderman & Tetlock, 1986; Sniderman, Tetlock, Carmines, & Peterson, 1993; Tetlock, 1994). Adherence to traditional values—without concomitant racial prejudice—could drive Whites' responses to SR measures and their opinions on racial policy issues. For example, Whites' devotion to true equality may lead them to oppose what they might view as inherently inequitable policies, such as affirmative action, because it provides advantages for some social groups and not others. Similarly affirmative action may be perceived to violate the traditional principle of judging people on their merits, not their skin color. Consequently, opposition to such policies may result from their perceived violation of widely and closely held principles rather than racism.
A second possibility is that the observed associations of SR with policy preferences may be spurious because both may reflect preferences about the scope and character of government interventions in private life. Liberals often argue that government should help people in need of economic help, in addition to helping its citizens to thrive in a variety of other ways. In contrast, conservatives often argue that government should allow citizens to get ahead economically on their own and should interfere with the free enterprise system as little as possible. If a person prefers a small government that carries out a very limited set of activities and minimizes its intrusions into the lives of its citizens, this preference could inspire opposition to policies that would help Blacks or any other groups. And such general preferences about the size of government may also determine responses to SR measures that ask about government action directly. As a result, what appears to be an effect of SR on policy attitudes may occur because both are caused by preferences about size of government. Consequently, SR may predict racial policy stances because measures of SR force people who endorse conservative views of what government should do to agree with the SR items (Sniderman & Tetlock, 1986; Sniderman et al., 1991; Tetlock, 1994; also see Roth, 1990).
These alternative interpretations are reasonable, and much research has been done to address them. To overcome the potential problems of spuriousness outlined above, previous studies have typically conducted multiple regressions to gauge the ability of SR to predict policy attitudes or candidate evaluations while holding constant various potentially confounding variables, including race-neutral conservatism, beliefs about the proper size of government, and/or egalitarian values. And in such analyses, SR has usually continued to predict attitudes when statistically controlling for these competing constructs (see, e.g., Hughes, 1997; Kinder & Sanders, 1996; Sears, 1988; Sears et al., 1997; also see Henry & Sears, 2002; Sears & Henry, 2003, 2005). Generally, though, these studies have not controlled for all of the variables that we include in the structural equation models reported below.
Given the popularity of the SR construct, it is surprising that there remain some unanswered, fundamental questions. Many past studies have identified outcome variables of which SR is a predictor (e.g., Henry & Sears, 2002; Hughes, 1997; Kinder & Sanders, 1996; Sears, 1988; Sears et al., 1997), but few have pointed out the instances in which SR does not predict. Another way to evaluate the performance of SR in predicting policy preferences is to assess discriminant validity, by examining whether SR predicts policy attitudes that do not involve a focus on Black Americans (see Schuman, 2000). If Black-focused SR truly represents racism, it should work well in predicting Black-targeted policy attitudes and less well in predicting other, non-Black-targeted policy attitudes—even if these non-Black-targeted policies still represent government-sponsored redistribution of wealth. However, if SR's association with Black-focused policy attitudes is spuriously attributable to ideological conservatism or beliefs about government's proper size or role, then SR should be an equally effective predictor of attitudes toward a wide variety of redistributive policies, regardless of their focus. Thus, the claim that SR is a measure of Black-focused racial prejudice can be tested by attempting to identify the limiting conditions of its predictive abilities.
To our knowledge, three attempts have been made to examine SR's ability to predict a range of policy attitudes across racial and nonracial domains. Kinder and Sanders (1996) found that SR was a better predictor of racial policy attitudes (e.g., integration of schools, preferential hiring of Blacks, and racial quotas in college admissions) than of nonracial policies, such as gay rights and defense spending. In addition, SR predicted attitudes toward “issues where race is present only by assumption” (e.g., food stamps and sanctions against the former apartheid regime in South Africa; Kinder & Sanders, 1996, p. 121). These findings are consistent with the conclusion that the SR scale does not predict opposition to all government activities equally well. Similarly, Kinder and Mendelberg (2000) documented stronger apparent effects of racial prejudice on aid to Blacks and other race-targeted policies than on social welfare policies not targeted for blacks (while controlling for individualism and egalitarianism values). Sears, Citrin, Cheleden, & van Laar (1999) found that SR predicted opposition to explicitly Black-targeted policies better than it did opposition to implicitly Black-targeted policies (e.g., crime or welfare) and in turn better than opposition to policies explicitly targeted for other groups (e.g., immigration or language policies).
The studies reported below sought to conduct analyses employing the best facets of all of these previous studies while improving the methods employed. We conducted covariance structure modeling, held constant all control variables at once, and gauged the ability of SR to predict attitudes toward a variety of policies, both racial and nonracial. Specifically, we assessed whether:
The first study involved analysis of data from two national samples of adults who were interviewed for the American National Election Studies (ANES). To identify what policies SR does and does not predict, we used the ANES data and structural equation models to examine SR's associations, controlling for other factors, with a variety of policy attitudes—some racial, some nonracial. In the second study, we experimentally manipulated the beneficiaries of a policy (Blacks vs. women) to assess whether removing race targeting weakened the ability of SR to predict preferences among college students.
For the 1990 American National Election Study (Miller, Kinder, Rosenstone, & the National Election Studies, 1999), a representative national sample of 1,980 adults was interviewed face-to-face in their homes between November 6, 1990, and January 26, 1991, following the U.S. Presidential Election. The response rate was 71.4%.
The 2000 edition of the American National Election Study (The National Election Studies, 2001) consisted of both preelection and postelection interviews, which were conducted either by telephone or face-to-face. A representative national sample of 1,807 adults were interviewed before the election (i.e., between September 5 and November 6), and 1,555 were interviewed after the election (i.e., between November 8 and December 18). The overall response rate was 61.2%. Some of the variables we used were assessed during the preelection interview, whereas others were asked during the postelection interview. Specifically, all of the policy items, thermometer ratings of Republicans and Democrats, party identification, and liberal-conservative self-identification were assessed in the preelection interview. The rest of the questions, including SR and thermometer ratings of conservatives and liberals, were asked after the election.
The NES symbolic racism scale consisted of four items, which asked respondents to agree or disagree with the following statements: “Over the past few years, Blacks have gotten less than they deserve” (SR1); “Irish, Italians, Jewish and many other minorities overcame prejudice and worked their way up. Blacks should do the same without any special favors” (SR 2); “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” (SR 3); and “Generations of slavery and discrimination have created conditions that make it difficult for Blacks to work their way out of the lower class” (reversed coded; SR4). Responses could range across five values, which were coded between 0 (disagree strongly) and 1 (agree strongly).
Anti-egalitarian values were measured with four questions that asked about what the nation ought to do about social equality. They offered the same response choices as the SR items. The statements were: “Our society should do whatever is necessary to make sure that everyone has equal opportunity to succeed” (reverse coded; Anti-egalitarianism 1), “This country would be better off if we worried less about how equal people are” (Anti-egalitarianism 2), “It is not really that big a problem if some people have more of a chance in life than others” (Anti-egalitarianism 3), and “If people were treated more equally in this country, we would have many fewer problems” (reverse coded; Anti-egalitarianism 4).
These items were coded so that higher scores indicated greater anti-egalitarianism.
Liberal-conservative ideology was assessed with two measures: (a) liberal-conservative self-identification on a 7-point scale, and (b) the difference between ratings of liberals and conservatives of a 101-point feeling thermometer (assessing how warm or cold respondents felt toward each group). Scores were coded to range from 0 to 1, with higher values indicating greater conservatism. Many respondents responded “don't know” or “I haven't thought much about this” when asked to place themselves on the 7-point liberal-to-conservative continuum. These people were asked a follow-up: “If you had to choose, would you consider yourself a liberal or a conservative?” People who said “liberal” were coded as being slight liberals (i.e., a score of 0.33), and people who said “conservative” were coded as being slight conservatives (i.e., 0.67). People who replied “moderate or middle of the road” to the follow-up question were coded at the midpoint of the scale.
Two items measured respondents' opinions about the proper role and size of government. Respondents chose one of two statements that most agreed with their own views. One question (Desired govt. size 1) asked respondents to choose between (a) “the less government the better,” or (b) “there are more things that government should be doing.” The second item (Desired govt. size 2) offered these statements: (a) “We need a strong government to handle today's complex economic problems,” or (b) “the free market can handle these problems without government being involved.” Higher scores (on a 0-to-1 scale) indicated a desire for less government.
Party identification was measured in two ways. First, respondents placed themselves on a 7-point scale from strong Democrat to strong Republican. Second, we computed the difference between the respondents' thermometer ratings of Republicans and of Democrats. In both cases, the recoded scores ranged from 0 to 1, with higher scores reflecting greater identification with the Republican Party.1
The following question measured attitudes toward affirmative action in both studies: “Some people say that, because of past discrimination, Blacks should be given preference in hiring and promotion. Others say that such preference in hiring and promotion. Others say that such preference in hiring and promotion of Blacks is wrong because it gives Blacks advantages they haven't earned. What about your opinion—are you for or against preferential hiring of Blacks?” Responses had four possible values: 0 (favor strongly), 0.25 (favor not strongly), 0.75 (oppose not strongly), and 1 (oppose strongly).
The 2000 NES also included a separate assessment of attitudes toward the government requiring companies to adopt affirmative action to compensate for previous discriminatory policies against Blacks (see Stoker, 1998). It read: “Some people think that if a company has a history of discriminating against Blacks when making hiring decisions, then they should be required to have an affirmative action program that gives Blacks preference in hiring. What do you think? Should companies that have discriminated against Blacks have to have an affirmative action program? Do you feel strongly or not strongly?” Responses to this question also ranged on a 4-point scale from 0 (Feel strongly—should have to have affirmative action) to 1 (Feel strongly—should not have to have affirmative action).
Attitudes toward federal spending on food stamps (used in both studies), welfare (Study 2), and social security (both studies), as well as attitudes toward assistance to the homeless (Study 1) and assistance to the poor (Study 2), were measured by items that asked respondents whether they believed government spending on these programs should be increased (coded 0), decreased (coded 1), or kept the same (coded 0.5). (Those few participants who volunteered a response to any of the spending questions that the program should be completely eliminated were coded as supporting decreased funding.)
Age, gender, level of highest educational attainment, and annual household income were included as demographic control variables.
We restricted our analyses to those respondents who identified their race as “White.” Although all respondents were asked most of the questions we used, the egalitarianism questions and the preferential hiring question were asked only of a randomly selected subset of respondents. In our case, this limited the data to 825 White respondents. Of these individuals, only respondents who provided valid values on all variables were included in the analyses reported below. As a consequence, 256 cases were excluded, and our final sample size was 569.2 Generally, missing values made up less than 1% of each variable's responses. (The item asking for self-identification on the liberal-conservative dimension had the most missing data [8.6% of cases].) Fifty-one percent of the sample was female. The median respondent was 43 years old and, in terms of education completed, had a high school diploma with some college courses completed. The median family income was in the $25,000 to $30,000 category (in 1989 dollars).
The data came from 708 respondents to the 2000 NES who identified their race as being “White” and who had complete data. (There were originally 1,393 White respondents.3) The sample was similar demographically to the sample of Study 1. Females and males were evenly split (i.e., women represented 50.5% of the sample), and the mean age was 47 years. The average pretax, household income was between $50,000 and $60,000 (in 1999 dollars).
For each sample, a covariance structure model was specified in which the policy attitudes (each presumed to be perfectly measured with a single indicator) were predicted by latent factors representing symbolic racism, egalitarianism, liberal-conservative ideology, desired size of government, and party identification (each tapped by multiple indicators), as well as the demographics (each presumed to be perfectly measured with a single indicator). The predictors were allowed to freely covary with one another, and the errors to the policy attitudes were also allowed to covary freely with one another. The metric of each latent variable was specified by fixing an indicator's loading at 1.0, thus allowing us to calculate unstandardized regression coefficients using the robust maximum likelihood technique in EQS 5.7b (Bentler, 1992).
Descriptive statistics and the factor loadings for the items are reported in Table I. The model for Sample 1 fit the observed data fairly well, χ2 (139) = 441.1, p < .001, χ2/df = 3.2, CFI = .89, AGFI = .87, RMSEA = .06. As shown in Table 2, egalitarianism, liberal-conservative ideology, and party identification did not have reliable independent effects on respondents' attitudes toward preferential hiring and promotion of Black Americans. Desired government size did predict policy attitudes as expected: people who desired a smaller, less activist government were more likely to oppose affirmative action (b = .16, p < .001). SR has a significant and sizable effect, such that people scoring high were especially likely to oppose the preferential hiring and promotion of Blacks (b = .53, p < .001).
A model similar to the unconstrained model specified with Sample 1 was tested with the 2000 data. Table 3 provides the path coefficients between each policy item and the latent predictor variables. The model fit the data well, χ2 (157) = 387.2, p < .001, χ2/df = 2.5, CFI = .95, AGFI = .91, RMSEA = .05. In predicting opposition to the preferential hiring and promotion of Blacks, SR's effect was slightly larger than it was in the 1990 survey (b = .59, p < .001). None of the other predictors had a statistically significant contribution to predicting this policy. People's attitudes toward enforced affirmative action as compensation for past discrimination were also closely tied to SR (b = .66, p < .001). In this case, however, desired government size also had an independent effect (b = .25, p < .01).
As shown in Table 2, people who scored high on symbolic racism were not significantly more opposed to any of the nonracial policy attitudes. The effect of SR was nonsignificant on attitudes regarding food stamps and aid to the homeless, and people who scored high on SR were more likely to favor increased spending on social security. We tested whether or not the slopes predicting affirmative action differed from the slopes predicting the other policies by specifying models in which pairs of slopes were constrained to equality and compared these to the unconstrained model described above. For example, to test whether SR was indeed better at predicting attitudes toward affirmative action than attitudes toward aid to the homeless, we constrained the SR-to-affirmative-action and the SR-to-aid-to-homeless paths to equality. Doing this harmed the overall fit of the model, Δχ2 (1) = 37.18, p < .0001, so the difference between the slopes predicting affirmative action and aid to the homeless is statistically significant. Similarly, the difference between SR's prediction of affirmative action and food stamps differed significantly, Δχ2 (1) = 7.12, p < .01.
Although welfare may frequently be interpreted as a racial policy (Gilens, 1999), the NES item does not specifically mention racial categories, so it is not as explicitly focused on race as the two affirmative action items. As can be seen in Table 3, SR predicted opposition to welfare to a moderately strong degree (b = .38, p < .01). Again, we tested two models in which the paths from SR to preferential treatment and, separately, to compensatory affirmative action were constrained to be equal to the slope predicting welfare attitudes in the 2000 data. These models revealed that the welfare coefficient does not differ significantly from the coefficients for preferential treatment, Δχ2 (1) = 2.26, p > .10, or for compensatory affirmative action, Δχ2 (1) = 2.35, p > .10. The contribution made by attitudes about government size (b = .20, p < .01) is statistically significant. In the case of welfare, the effect of attitudes toward limited government services is more similar to the contribution of SR. Egalitarianism, party identification, and liberal-conservative ideology had no independent effects on attitudes toward spending on welfare.
The effect of SR on attitudes toward spending on food stamps was larger than in Sample 1 and was statistically significant (b = .28, p < .05), but it was smaller than SR's effects in predicting affirmative action or welfare attitudes. None of the other predictor variables included in the model contributed to a statistically significant degree. The constrained models indicated that SR was a better predictor of both forms of affirmative action than food stamps, Δχ2 (1) = 5.19, p < .05 and Δχ2 (1) = 4.34, p < .05, for preferential treatment and for compensatory affirmative action, respectively.
SR did not show a statistically significant relationship with attitudes regarding spending on aid to the poor (b = −.12, ns). In this case, both anti-egalitarianism (b = .51, p < .001) and attitudes toward government size (b = .20, p < .01) were the only other variables that exhibited statistically significant effects. The constrained models revealed that not only did SR's prediction of aid to the poor differ significantly from the slopes for the two forms of affirmative action, but it also differed from the slope predicting welfare, Δχ2 (1) = 14.28, p < .001.
The findings concerning social security replicated what was found in Sample 1. SR negatively predicted opposition to social security spending (b = −.35, p < .001). Attitudes toward government size (b = .20, p < .01) and anti-egalitarianism (b = .34, p < .01) had positive effects, as they did in 1990; however, party identification did not have an independent effect in 2000.
We found evidence consistent with the claim that attitudes toward redistributive policies are driven in part by egalitarian values, attitudes toward big government, and liberal-conservative ideology. But even when controlling for these factors, SR was strongly related to attitudes toward affirmative action for Black Americans. And SR was not positively associated with opposition to redistributive government efforts that did not explicitly target Blacks. SR did not predict Whites' attitudes toward assistance to the homeless or toward government spending on food stamps. This pattern is consistent with the conclusion that SR may be a determinant of opposition to government aid to Blacks via the processes outlined by SR theory. Welfare spending appears to be an exception—the test of nested models revealed that SR predicted welfare attitudes to the same degree as the affirmative action programs—but the magnitude of the slope puts it somewhere between affirmative action programs and spending on food stamps and the poor in terms of SR's ability to predict variability on this issue. This may not be too surprising because welfare in particular carries a clear racial connotation for most Americans (Gilens, 1999). The greater success of SR in predicting race-targeted than nontargeted redistributive programs is consistent with the claim that SR is not merely a proxy for general political conservatism, opposition to economic redistribution, or preference for small government. Rather, race appears to be an important factor (although certainly not the only one) in the minds of Whites when they calculate their stance on racial policies such as affirmative action. The finding that SR predicts attitudes toward affirmative action more strongly than attitudes toward food stamps or aid to the poor—is consistent with the contention that symbolic racism is tapping some sort of race-based sentiment.
Furthermore, SR's correlation with support for increased governmental spending on social security goes against the criticism that SR is just a nonracial measure of conservatism and/or preference for smaller government. A possible explanation for this correlation is that social security may have particularly racial connotations for Americans. However, as Winter (2006) argued, for many Americans the prototypical social security recipient is a White person. If so, SR's relation with endorsement of social security spending may have occurred because people high in SR are especially likely to endorse government programs that offer help to older Whites who have retired after years of hard work. These data replicate the correlations between SR and social security spending found by Winter (2006). Although the result was not expected, this interpretation is consistent with the essentially racial nature of SR.
Another explanation for SR's differential predictive abilities could be that White respondents who score high on the SR scale are troubled by the idea of government helping Blacks, but they also have a problem with the idea of government helping any targeted social group. Perhaps SR predicts opposition to preferential treatment of Blacks better than it predicts attitudes toward food stamps because beneficiaries of affirmative action are chosen based solely on their membership in a particular group and not because of any demonstrated socioeconomic need. According to this argument, therefore, SR should predict any policy attitude where any targeted groups stand to benefit, not just racial groups (see Carmines & Merriman, 1993). The fact that the affirmative action variables in Study 1 specifically mentioned Blacks would be of secondary importance; high-SR respondents would have opposed this policy had any group in society been singled out for special treatment. It would be worthwhile, then, to compare Black-targeted redistributive policies to policies targeting other groups. If this group-targeting argument is correct, high SR should correlate with opposition to redistributive policies whenever a particular social group stands to benefit (also see Sears & Henry, 2003).
In Study 2, the beneficiary of the policy was experimentally manipulated by asking respondents how supportive they were of the government setting aside contracts for either (1) Black-owned businesses or (2) women-owned businesses. Bobo and Kluegel (1993) used a similar strategy (also see Sidanius, Singh, Hetts, & Federico, 2000; Stoker, 1998) and found that prejudice predicted Whites' opinions about race-targeted policies better than they predicted opinions about income-targeted versions of the same policies. This is consistent with the expectations of symbolic racism theory, but Bobo and Kluegel did not measure symbolic racism. Furthermore, our distinction is not between race- and income-targeted policies; rather, we explore whether SR differentially predicts attitudes toward the same policy that targets different groups in society. In other words, we used an experimental design to test whether the relationship between symbolic racism and attitudes toward affirmative action is moderated by characteristics of the groups who stand to benefit.
A 13-page questionnaire concerning a variety of social and political issues was administered to 210 undergraduate students at California State University, Northridge, during the Fall, 2001, and Spring, 2002, semesters. Students received partial credit for the requirements of an introductory psychology course. Only the 84 students who reported their ethnicity/race as being “White/Caucasian” were eligible for inclusion in this study. Of these, 69% were female, and 31% were male. Seven cases with missing data were removed, which created a final sample size of 77.
Six items assessed SR. The item wordings are provided in the appendix. The response options ranged along a 7-point scale. As with all of the measured variables reported below, the variable scores were transformed to 0-to-1 scales. These six items were averaged to create a single composite SR score for each respondent (Cronbach's alpha = .78).
Twelve items assessed the value people placed on equality of opportunity and equality of outcome (see McClosky & Zaller, 1984; Rae, 1981). A sample item from the former scale is, “How much do you think it would help or hurt America if we were to provide equal opportunities for all social groups to get ahead in life?” An example from the outcome scale is, “Do you think reducing the differences in wealth between people in America is a good thing, a bad thing, or neither good nor bad?” There were seven response options that were coded on a 0-to-1 scale. Higher scores indicated greater anti-egalitarianism. The resulting factors had acceptable to fairly good reliability (α = .58 for outcomes and α = .73 for opportunity).
Liberal-conservatism was measured on a 7-point scale ranging from 0 (extremely liberal) to 1 (extremely conservative).
Respondents were asked to identify as Democrat, Republican, or Independent or something else. Those who considered themselves Democrats or Republicans were asked whether they considered their affiliation to be strong or not very strong, and those who selected Independent or something else were asked whether they thought of themselves as being closer to one party or the other or equally close to both parties. All of these response options created a 7-point scale from 0 (strong Democrat) to 1 (strong Republican).
Feelings toward Democrats and Republicans were rated on a 9-point scale from extremely warm to extremely cold. The difference between each respondent's ratings was transformed to range from 0 (i.e., maximally warm toward Democrats) to 1 (i.e., maximally warm toward Republicans). The mean of these two items defined respondents' political party preference scores. The reliability of these questions was quite high (α = .79).
Four items measured individuals' beliefs about the proper role of government in the lives of individual citizens and businesses. Participants were asked questions, such as: “Do you think government involvement in the lives of individual people is good, bad, or neither good nor bad?” Responses were made on 7-point scales (transformed to 0-to-1 scales); higher scores indicated support for a more limited role for government. The reliability for the four items was acceptable (α = .58).
The dependent variable was attitude toward government contract set-asides. The target group of the policy was manipulated between groups. Randomly selected halves of the sample were asked about set-asides for businesses owned by either Black Americans or women. The exact wording was as follows (with the wording manipulations presented in parentheses): “Some people think that government agencies should definitely give some of their contracts to (women-owned/Black-owned) businesses. Other people think that government agencies should not use the (gender/race) of the owners of a business when deciding whether to give a contract to it. How about your opinion–do you favor or oppose government agencies definitely giving some contracts to (women-owned/Black-owned) businesses?” Participants rated the degree to which they favored or opposed this practice on a 7-point scale from 0 (strongly favor) to 1 (strongly oppose). The manipulated policy beneficiary variable was coded 0 for those asked about Black-owned businesses and 1 for those asked about women-owned businesses.
The key test is the comparison between the SR coefficients predicting scores on the two versions of the contract set-asides item. If SR mainly reflects racism, and not simply concerns about group-targeted policies, then the simple slope for the Black-targeted policy item should be greater than the simple slope for the women-targeted item. We conducted a hierarchical multiple regression analysis to see whether the ability of SR to predict policy opposition depended on the beneficiary of the policy. To test this SR × Policy Beneficiary interaction, we centered the scores around the respective variables' means and created an interaction term by multiplying participants' SR score by their score for the dummy-coded policy-beneficiary manipulation (see Table 4 for descriptive statistics and item correlations). In step 1 of the multiple regression analysis, the policy beneficiary variable and all of the control variables (viz., liberal-conservative self-identification, political party preference, limited government role, egalitarianism of opportunity, and egalitarianism of outcomes) were entered as predictors of policy opposition; at step 2, SR was added; and, finally the SR × Policy Beneficiary term was added in step 3.
The results of the multiple regression analysis are presented in Table 5. Step 1 does not do a particularly good job of predicting policy opposition. However, consistent with the previous studies, the addition of symbolic racism in the second step improves the regression model to a statistically significant degree. Higher scores on SR were associated with greater opposition to contract set-asides, b = 0.46, p < .05. More important, in the final model specification, there is evidence for a Symbolic Racism × Policy Beneficiary interaction, b = −0.75, p < .05. Adding this interaction term to the overall equation significantly improves the prediction of the model over and above the contributions of all other variables, ΔR2 = .05, p < .05.
To interpret the interaction, we calculated simple slopes for both the Black- beneficiaries and the female-beneficiaries subsamples (see Aiken & West, 1991; Preacher, Curran, & Bauer, 2006). The mean values of the control variables (i.e., values of 0, because the variables were centered) were plugged into the rearranged equations. The simple slopes analysis revealed that SR did not significantly predict opposition to contract set-asides when the beneficiaries were women, b = 0.12, SE b = 0.26, p = .64, but strongly predicted opposition when the beneficiaries were Blacks, b = 0.87, SE b = 0.28, p = .003.4
The results of Study 2 bolster the findings from the previous study. As with the studies that used national probability samples, the student data provide an opportunity to test the effect of SR while controlling for the effects of other, race-neutral variables. This experimental manipulation affords the additional opportunity to test SR's independent effect on attitudes toward two versions of the same policy. The only difference between the two versions of affirmative action was the target group that was specified as the beneficiary of the policy: Black-owned businesses versus women-owned businesses.
SR predicted attitudes toward race-focused affirmative action but not attitudes toward gender-focused affirmative action. The pattern of slopes, in combination with the findings of the other study that used different measures and vastly different samples, represents evidence that symbolic racism taps racial sentiment, not just respondents' frustration with or anger over redistributive policies or group-targeted policies in general. These results are consistent with the hypothesis that White opposition to racial policies incorporates an explicitly racial component; opposition does not reflect a blanket rejection of group-based targeting.
The studies presented here provide evidence that, although Whites' attitudes toward racial policies may be the results of general political orientations, these attitudes are also a function of considerations specific to race. First, our studies explore the possibility that measures of symbolic racism (SR) have sometimes posed a no-win situation for nonracist, conservative respondents (Sniderman, Crosby, & Howell, 2000; Tetlock, 1994; also see Feldman & Huddy, 2005). Such individuals, when answering SR questions, would agree that Blacks should work their way up just like everyone else. But these individuals may harbor no ill will toward Black Americans in particular; they may simply feel that everyone should have to work his or her way up. The SR scale, therefore, may confound anti-Black racism with basic race-neutral, conservative, and individualistic principles.
The studies presented here appear to suggest that this account does not completely explain the covariations observed in past studies. Conservative ideology and opposition to large government were often strongly associated with policy preferences and with symbolic racism. However, the SR scale strongly predicted Whites' opposition to policies designed to aid Blacks, even after the scale's association with ideology and beliefs about government involvement were statistically controlled. This pattern appeared in each instance when attitudes toward a race-focused policy were being predicted.
SR's predictive properties become attenuated when the policy strayed from specifically targeting African Americans. When attitudes such as ideology and the role of government were controlled, symbolic racism did not predict attitudes toward assistance to the homeless, assistance to the poor, or food stamps and predicted attitudes toward welfare as powerfully as it predicted attitudes toward affirmative action for Black Americans. This is despite the possibility that the homeless, the poor, and recipients of food stamps and welfare may be viewed by many Whites as being disproportionately Black (Gilens, 1999).
Our data also challenge the conclusion that the SR scale's apparent predictive success is attributable partly to the fact its items are too closely related in content to the racial policy attitudes being predicted. Study 2 used variations of the SR scale that are not as problematic in this regard as were the NES scales, which were more closely tied to denial of anti-Black discrimination and resentment of Blacks' anger over their position in society than to perceptions of undeserved group-based claims or economic redistribution. And in Study 2, SR exhibited stronger associations with race-targeted policy stances than with gender-targeted policy stances despite identical wordings of the SR questions.
Our findings in this regard resonate with findings reported by Sears and his colleagues, who found that SR predicts opposition to Black candidates for public office independently of liberal-conservatism (Sears et al., 1997; Tarman & Sears, 2005). This has also been more recently demonstrated with the latest version of the symbolic racism scale (Henry & Sears, 2002). Our study and other recent research reveal that the effects of SR on racial policy preferences are almost identical irrespective of what particular SR items are used (e.g., Tarman & Sears, 2005). Most important, the effects are the same regardless of whether the items refer to politics and government or merely to Blacks' social status, such as the magnitude of discrimination they face or how hard-working they are perceived to be.
Our findings also resonate with other past studies. In particular, a number of studies have controlled for individuals' nonracial attitudes (e.g., liberal-conservatism), and doing so did not weaken the associations of SR with racial policy attitudes (see, e.g., Kinder & Sanders, 1996; Sears et al., 1997; Tarman & Sears, 2005; also see Hutchings & Valentino, 2004; Sears & Henry, 2005). The complementary (but less frequently adopted) paradigm used here is to compare SR's impact on attitudes toward explicitly racial, implicitly racial, and nonracial policies.
Much evidence has been presented before attesting to symbolic racism's predictive capabilities. But no past study has tested the discriminant validity of symbolic racism measures as we have here. By identifying the variables that it does predict and those that it does not, the results reported here suggest that much of the variance that explains opposition to explicitly racially targeted programs is due to anti-Black antipathy blended with perceptions of value violation, not simply anti-redistributive attitudes or attitudes toward group-based claims and demands.
Finally, these studies illustrate the continued importance and meaningfulness of parsing opinions about racial policies. Enough variability exists in White Americans' policy attitudes to make meaningful claims about psychological and political antecedents. Even attitudes toward affirmative action programs like preferences in hiring and promotion, which appear to be widely opposed, provide a fruitful context for exploring the influences of wide-ranging beliefs and psychological values. Indeed, affirmative action appears to be the most racial of all the redistributive policies examined here.
The first author was supported by a Grant-in-Aid Award from the Society for the Psychological Study of Social Issues while at the University of California, Los Angeles, and by a grant (T32-MH19728) from the National Institute of Mental Health while at The Ohio State University. Data collection at California State University, Northridge, was funded by grants from the College of Social and Behavioral Sciences and the Office of Graduate Studies, Research, and International Programs. Jon Krosnick is University Fellow at Resources for the Future. We thank Rosellen Bronson, Karina Kortbein, and Amber Landini for their help with data collection and entry. We also thank Peter Bentler and Chick Judd for statistical advice.
1We also tested a version of the model in which the two liberal-conservative ideology items and the two political party identification items all loaded on the same factor. The coefficients for the paths between the predictor and outcome variables in that model did not differ substantially from the model we report below. However, overall model fit was harmed.
2We also ran a model based on data with the missing values replaced. Using the expectationmaximization (EM) procedure available in EQS 6.1, we were able to keep the data from all 825 people. The magnitudes of the coefficients did not differ substantially from the results reported below.
3As with Sample 1, EM imputation of missing data in this sample yielded coefficients that were similar to the results reported below. However, because of the dramatic improvement in statistical power as a result of adding the missing 685 people, tests of significance generally had lower p-values.
4When we drop from the analyses all of those participants who belong to one of the beneficiary groups—in this case, that would entail excluding females—we are left with 26 male, White respondents. Despite the low statistical power among this sample, the SR × Policy Beneficiary interaction remains statistically significant, b = −2.23, SE b = 0.79, p = .01, and follows the expected pattern.
Joshua L. Rabinowitz, University of Michigan.
David O. Sears, University of California, Los Angeles.
Jim Sidanius, Harvard University.
Jon A. Krosnick, Stanford University.