Sophistication and the Antecedents of Whites' Racial Policy Attitudes - Racism, Ideology, and Affirmative Action in America

A number of researchers have argued that the effects of prejudice on the racial policy attitudes and general political beliefs of white Americans may be restricted to the poorly educated and politically unsophisticated. In contrast, rather than being motivated by prejudice, the racial policy attitudes and ideological values of the politically sophisticated white Americans should be more ﬁrmly informed and motivated by the tolerant values at the heart of American political culture. These values include such things as individualism, notions of fair play, and devotion to the principle of equality of opportunity. We tested this hypothesis using white respondents from the 1986 and 1992 National Election Studies. Our evidence generally indicated that racial policy attitudes and political ideology were more powerfully associated with ideologies of racial dominance and superiority among politically sophisticated white Americans than among political unsophisticated white Americans. Moreover, even among the sophisticated, we found that various forms of egalitarianism predicted support for—rather than opposition to—afﬁrmative action and that support for equal opportunity is not uniformly distributed across the political spectrum. Studies of the momentous changes in public opinion that accompanied the civil rights movement of the 1950s and 1960s have generally reached a common


Federico and Sidanius
conclusion about the trajectory of white racial attitudes over the last 40 years. While support for segregation and overt discrimination has given way to support for the formal principle of racial equality, this change has not been accompanied by increased white support for "race-targeted" policies aimed at implementing this principle, such as affirmative action (see Schuman et al. 1997;Sears 1988;Sears et al. 2000;Sniderman and Piazza 1993). Despite general consensus on this point, students of race relations in America have disagreed sharply about the origins of this "principle-implementation" gap.
On the one hand, consistent with the notion of an overall decline in racism among whites, a number of researchers have argued that white opposition to policies like affirmative action is now primarily a function of race-neutral political considerations, such as conservatism, individualism, and opposition to "big government," and that these considerations are themselves orthogonal to racist motives (Sniderman and Carmines 1997;Sniderman, Crosby, and Howell 2000;Sniderman and Piazza 1993). According to this perspective, whites oppose policies like affirmative action not out of a dislike for blacks or a desire to maintain their group's privileged position, but because they sincerely believe that such policies enshrine a regime of preferential treatment. In other words, it suggests that whites reject affirmative action for many of the same reasons they now oppose the sort of institutionalized discrimination dismantled by the civil rights movement: namely, because it violates key elements of what Gunnar Myrdal (1944) famously referred to as the "American Creed," such as equality before the law and equality of opportunity (see Connerly 1996), principles now thought to be consensually accepted throughout the mainstream political spectrum (Sniderman and Piazza 1993).
On the other hand, other researchers have suggested that racial prejudice and desires for group dominance continue to influence white attitudes toward these policies (e.g., Blumer 1961;Bobo 1988Gilens 1999;Jackman 1994;Sidanius and Pratto 1999;Sidanius, Pratto, and Bobo 1996;Sidanius et al. 2000;Turner, Singleton, and Musick 1984). This outlook suggests that a desire to preserve the white ingroup's interests and privileged position may still be responsible for opposition to policies aimed at implementing the principle of racial equality, even if racism is generally less prevalent than in the past. Moreover, while agreeing that support for equal opportunity has increased over the last few decades, these researchers also suggest that values of this sort are not endorsed to the same degree across the ideological spectrum, raising the possibility that disagreement about the value of formal equality may be more fundamental to American political culture than many have assumed (see Dawson 2000;Smith 1997).
While much of this debate has taken the form of a basic disagreement about the relative explanatory power of racial and race-neutral predispositions and the degree to which concern for formal equality has become evenly diffused throughout the political spectrum, an additional-and perhaps more interesting-disagreement about the effects of education on the relationships between these variables has also arisen. Consistent with a long line of studies indicating that education tends to promote tolerance (see Lipset 1960;McClosky and Zaller 1984;Steinberg and Selznick 1969), analyses suggesting a reduced role for prejudice in white racial policy attitudes have also argued that this should particularly be the case among educated whites. More precisely, several researchers have argued that education may attenuate the relationship between racial hostility and opposition to affirmative action, as well as the relationship between racism and various race-neutral predispositions (see . Since poorly educated individuals lack the expertise necessary for the comprehension and use of abstract political concepts, their racial policy attitudes-and more broadly, their general political outlook-should be more heavily colored by prejudiced considerations. In contrast, the knowledge possessed by well-educated individuals should allow them to bring complex ideological principles to bear on their racial policy attitudes (see Sniderman, Brody, and Tetlock 1991;. For example, in a review and endorsement of this approach, Jacoby writes, "Education also plays an important part in determining attitudes, but not simply because more education correlates with less racial prejudice: more educated whites are also more influenced by political ideas, thus amplifying the role of ideology and diminishing the relevance of race per se" (1994, p. 37; emphasis added). Thus, while racism and conservatism are perhaps not completely independent of one another, this correlation may be largely restricted to the poorly educated, who should be less aware of the actual principles underlying conservative ideology and the American Creed.
At a broader level, this line of reasoning also suggests that the increase in sophistication associated with schooling should make it easier for well-educated survey respondents to recognize the basic conflict between intolerant attitudes and the support for equal opportunity thought to be embodied in the American Creed-values believed to be accepted by liberals and conservatives alike at this point in history, given the decline in overt racism over the last few decades (see Sniderman, Tetlock, and Carmines 1993; see also McClosky and Zaller 1984;Zaller 1992). Thus, while the notion of consensual support for equal opportunity generally suggests a minimal relationship between ideology and concern for equal opportunity, it also suggests that this ideologically even pattern of support for formal equality may be most evident among the educated, who should be most aware of the fundamentally tolerant norms of American political culture. While investigators have not thoroughly examined the issue of whether concern for equal opportunity has become ideologically consensual, research has provided some evidence for the idea that education may reduce the degree to which prejudice is bound up with policy attitudes and generalized political predispositions. For example, Sniderman and his colleagues (e.g., Sniderman and Piazza 1993; have found that college-educated conservatives are no more likely than college-educated liberals to practice a double standard by offering greater support to government efforts to guarantee equal opportunity for women than to similar efforts targeted at blacks. In contrast, both poorly educated conservatives and poorly educated liberals tend to offer less support to black-targeted equal opportunity initiatives.
Despite these findings, other analyses have offered less support for the hypothesis that education should attenuate the effects of racism and have instead suggested that education may simply allow individuals to better align their racial policy preferences-and broader political outlook-with a desire to protect the dominant position of the in-group. In this vein, Sidanius and his colleagues (Sidanius et al. 1996; see also Sidanius and Pratto 1999) have found that education may strengthen the effects of racism and other group-relevant attitudes. In several samples of whites, they found that classical racism and group-based antiegalitarianism had a stronger relationship with both affirmative-action attitudes and generalized conservatism among the highly educated. Results such as these imply that the increases in political understanding associated with education may not lead whites to rely more heavily on race-neutral considerations when making policy judgments. In fact, in certain cases, they suggest that individuals with a better understanding of basic political concepts may find it easier to connect their attitudes toward group-relevant social policies with both the perceived interests of their own groups and their attitudes toward the out-groups the policies are designed to benefit (see Sidanius et al. 1996;Sidanius and Pratto 1999; see also Jackman and Muha 1984). 1 At a broader level, the existence of these relationships suggests that concern for equal opportunity-although stronger and more consensually expressed than in the past-is still not evenly distributed across the political spectrum, since conservatism may in part be rooted in a desire for group dominance (as suggested by a large body of research linking conservative attitudes to racism and antiegalitarianism; see Sidanius and Pratto 1993;Sidanius et al. 1996Sidanius et al. , 2000. Moreover, in line with the finding that education may actually strengthen the influence of antiegalitarian motives like racism, this may be particularly true of educated respondents, who are precisely those people most likely to understand both the hierarchy-enhancing implications of conservative ideology and the manner in which this ideology serves their competitive group interests. On the whole, then, existing evidence does not provide a clear pattern of support for the hypothesis that education should attenuate the effects of racial prejudice and produce a more consensual pattern of support for equal opportunity. However, a closer examination of the ways in which education is typically assumed to affect the relationship between prejudice, race-neutral 1. This is, of course, consistent with one of the more general conclusions reached by studies of attitude constraint and belief-system structure: i.e, the idea that sophisticated individuals should be more attuned to the implicational connections between different sets of beliefs (Converse 1964;Judd and Krosnick 1989;Lavine, Thomsen, and Gonzales 1997;Stimson 1975;Zaller 1992). values, and racial policy attitudes-along with recent work on the conceptualization and measurement of political understanding-suggests that the underlying content of this hypothesis may not have been adequately tested, by either its proponents or its critics. More precisely, most work on the effects of education on racial attitudes (e.g., Steinberg and Selznick 1969) implies that the effect of education on the organization of racial policy attitudes should be of a mediated rather than a direct character. That is, education should diminish the influence of racial prejudice primarily by virtue of its salutary effects on the learning and comprehension of abstract political concepts (cf. Bishop 1976;Stimson 1975; see also McClosky and Zaller 1984;Zaller 1990). By providing individuals with a better understanding of the ideas that animate mainstream political discourse, education is thought to facilitate the use of abstract ideological concepts and a deeper recognition of the basic incompatibility between racism and the norms of American political culture, thereby reducing the impact of purely prejudicial considerations. For example, Sniderman and Piazza (1993) highlight this mediated aspect of the effects of education when they note that "the better educated, by virtue of being better informed, more adept at picking up and processing information, and better able to handle abstract ideas, are more likely to have a grip on the official norms of the larger culture, which in the case of the United States prominently include tolerance" (p. 47; emphasis added).
In essence, then, hypotheses about the effects of education on the policy impact of prejudice would seem to focus most directly on the role of political sophistication, with education merely serving as an indicator of the latter. However, recent work has increasingly suggested that education may not be the most reliable measure of political sophistication. While education may very well provide individuals with the cognitive sophistication necessary for an advanced understanding of politics (e.g., Bishop 1976;Stimson 1975), it does not guarantee attention to and comprehension of political ideas, as suggested by the relatively low correlation between education and various measures of interest in politics, such as media exposure to political information (see Zaller 1990Zaller , 1992; see also Converse 1980). In light of this observation-and the shortcomings of other measures of political sophistication, such as political interest and involvement-a number of theorists have suggested that tests of political knowledge may provide the most accurate portrait of a given individual's level of political understanding (Fiske, Lau, and Smith 1990;Zaller 1990; see also Hamill, Lodge, and Blake 1985;Lau 1990;Lavine et al. 1997). Although there are a number of reasons for this expectation (see Fiske et al. 1990), perhaps the most important of these is inherent in the notion of factual political knowledge itself: as Zaller (1990) has observed, more than any other measure, knowledge "captures political learning that has actually occurred-that is, political ideas that have been encountered and comprehended and remain available for use" (p. 131). Education, while clearly connected to knowledge and awareness (Bishop 1976;Sniderman, Brody, and Tetlock 1991), may not in and of itself be as indicative of a genuine understanding of politics.
Empirical studies of the relative predictive utility of various political-sophistication measures have reinforced this general conclusion. One of the most comprehensive of these studies was conducted by Susan Fiske and her colleagues (Fiske et al. 1990), who looked at the relationship between a series of political-sophistication measures and advantages in the processing of political information. A factor analysis of the sophistication measures suggested five basic dimensions of political sophistication: political knowledge, political activity, electronic media use, print media use, and "political self-schematicity" (i.e., thinking of oneself as liberal or conservative). Further analyses revealed that, of these five factors, political knowledge was the most useful in predicting the sort of information-processing benefits suggested by theoretical models of political schematicity and understanding (see Fiske et al 1990;Lodge and Hamill 1986), such as reduced reading time for political materials, quicker decision making, and improved recall for relevant stimulus items. Perhaps even more importantly-given the focus of the present investigation-other analyses have suggested that political knowledge may also be a better index of the degree to which survey respondents are able to perceive implicational links among related ideas and use ideological considerations to structure their policy attitudes. In this vein, a number of studies have indicated that factual political knowledge is a stronger predictor of attitude constraint and response stability than education (Zaller 1990(Zaller , 1992; see also Converse 1980;Judd, Krosnick, and Milburn 1981).
Thus, a variety of evidence suggests that education might not be the most reliable indicator of the degree to which people truly understand abstract political concepts. As such, the fact that virtually all of the studies in the present area of inquiry have relied on education as an index of political understanding implies that some of the competing hypotheses in this area may not have been tested as thoroughly and directly as possible. More precisely, if it is true that educational attainment is not the most accurate index of political understanding, then these studies may not have clearly identified those respondents who were in the best position to bring purely political considerations to bear on their racial policy reasoning and recognize the essential contradiction between racially motivated thinking and the consensually accepted principles of the American Creed (cf. . Given this complication, we felt that a reexamination of these predictions-using a more direct indicator of respondents' understanding of abstract political concepts-was in order. Moreover, in light of the aforementioned confusion about whether support for formal equality has truly become consensual, we also decided to examine the relationships between egalitarianism, racial policy attitudes, and ideology, with a special focus on the attitudes of sophisticates.

Overview of the Analyses
In order to address these concerns, we relied on data provided by white respondents from the 1986 and 1992 American National Election Studies (NES). Like most of the major NES surveys, the instruments used during each of these years contained a number of specific political-knowledge measures, as well as other items from which additional knowledge measures may be constructed. Moreover, each survey contained several items measuring racial prejudice, racial policy attitudes, and various race-neutral political values (such as conservatism, individualism, and beliefs about government). Using these measures, we set out to address four primary questions: 1. After the effects of conservatism and other race-neutral political values are taken into account, does political sophistication attenuate or strengthen the relationship between racial prejudice and opposition to affirmative action? 2. Does an advanced understanding of abstract political concepts increase or decrease the magnitude of the relationship between political conservatism and racial prejudice? 3. Is opposition to affirmative action driven by a genuine concern for equality, and is the strength of the relationship increased as a function of political sophistication? 4. How consensual a norm is concern for equal opportunity in American political culture? Is this concern distributed evenly across the political spectrum? Is it more consensual among the sophisticated?

Data and Measures
As noted above, our data were taken from the 1986 ( ) and 1992 N p 2,176 ( ) NES. In 1986, only respondents who received form B of that N p 2,487 year's survey were asked the necessary battery of racial attitude items. Therefore, only the white respondents who received this form ( ) were N p 922 included in the analyses. In 1992, all respondents received the necessary items, so the full sample of white respondents ( ) was used. 2 Items from N p 1,880 2. In 1986, for form B, the response rate was 67.9 percent. In 1992, the response rate for the preelection survey was 74 percent for respondents included as part of the 1992 NES cross-section and 77.7 percent for respondents included as part of the 1990-92 NES panel study; the reinterview rate for the postelection survey was 89.3 percent for respondents included as part of the 1992 NES cross-section, and 85 percent for respondents included as part of the 1990-92 NES panel study. Also, in the analyses that follow, further reductions in sample size occurred because of missing data for some variables. Final sample sizes for each analysis are shown in tables 1-4. It should be noted that most of these additional reductions in sample size were due to missing values for key variables rather than demographics. For example, the variables with the largest numbers of missing values were beliefs about government (105 missing) and opposition to affirmative action (107 missing) in 1986, and evaluative racial superiority (223 missing) and these surveys were used to operationalize five general clusters of variables: (1) political sophistication, (2) race-neutral political values, (3) racial prejudice, (4) general egalitarianism, and (5) and opposition to affirmative action. With the exception of the demographic measures described below, all of the measures described below were recoded to run from zero to one. Further information on the measures from each survey is provided in the appendix.

political sophistication
Based on research suggesting that accurate political knowledge is the strongest predictor of "expert" patterns of political cognition and belief organization (see above; see also Fiske et al. 1990;Zaller 1990), we operationalized political sophistication in terms of respondents' performance on a series of political knowledge items. In the 1986 NES, 13 indicators were used. Six of these asked respondents to identify the job or political office held by a series of political figures, while two additional questions asked respondents which party had the most members in each of the two houses of Congress before the 1986 election. For each of these eight items, respondents were given a score of one for a correct response and a score of zero for an incorrect response. Five additional indicators of respondents' political knowledge were generated by comparing their placements of five pairs of "opposed" political figures and parties on standard 7-point policy scales (cf. Zaller 1990). In order to be given a correct score on each pair, respondents had to locate the two groups or candidates in the correct left-right order and at least two scale units apart. Correct answers were given a score of one, and incorrect answers were given a score of zero. In the 1992 NES, 10 indicators were used. Four of these asked respondents to identify the job or office held by a series of political figures, while three additional items asked about various factual aspects of the political system (e.g., who nominates judges to the federal courts, which branch of government evaluates the constitutionality of laws, which party is more conservative, etc.). Three more indices were created by comparing respondents' placements of the two political parties on defense, government spending, and generalized liberalism-conservatism. Again, respondents had to place the two parties in the correct left-right order and at least two scale units apart to be given a correct score. These items were scored just like they were in 1986.
Once the items were scored, overall measures of political sophistication for each data set were then generated by adding respondents' scores on all 13 items in the 1986 NES and all 10 items in the 1992 NES. Both sets of sophistication measures scaled reliably ( in 1986; in 1992

race-neutral political values
Political conservatism. In each data set, a composite index of respondents' political beliefs was created by averaging respondents' scores on four different measures: (1) respondents' self-placement on a 7-point ideology scale, (2) the NES summary measure of partisan identification, (3) the difference between respondents' thermometer ratings of conservatives and liberals, and (4) the difference between respondents' thermometer ratings of the Republican Party and the Democratic Party. Before each scale was created, each item was rescaled on a zero to one interval. Scores were coded such that higher values indicated a stronger tilt to the right. Together, the four scores formed reliable scales in each data set ( in 1986; in 1992). a p .80 a p .85 Beliefs about government. This variable, which measures respondents' level of support for government intervention in social and economic matters, was included as a second index of respondents' political values. In both surveys, it was measured using a single indicator: the standard 7-point NES fewerversus-more government services and spending item. These items were coded such that higher scores indicated a preference for less government.
Individualism. In the 1986 NES, an additional six-item measure of raceneutral political values was created using the economic individualism scale developed by Feldman (1988) and employed in several subsequent studies of racial attitudes (e.g., Sears, Henry, and Kosterman 2000). In general, these items measure the degree to which respondents endorsed beliefs associated with the Protestant work ethic, such as hard work, self-reliance, and the idea that one's social outcomes are primarily a function of individual effort. Each item was coded such that higher scores indicated a greater endorsement of individualistic values ( ). This measure was not available in the 1992 a p .60 NES.

measures of racial prejudice
In order to gauge the effects of respondents' basic racial attitudes on opposition to affirmative action, we measured prejudice as a sense of white in-group superiority in both data sets. 3 In the 1986 NES, prejudice was measured in terms of classical racism. This variable measured the degree to which respondents believed that blacks are "naturally" inferior to whites (see Sidanius et al. 1996; see also Sears 1988). It was assessed by asking whether respondents believed that black disadvantage was due to either (1) an innate lack of ability or (2) a "divine" plan. Higher scores indicated greater racism ( ). In the 1992 NES, racial prejudice was indexed in terms of eval-a p .52 uative racial superiority. This index-which measured a slightly less extreme form of belief in the inferiority of blacks-was created using items that asked respondents to rate whites and blacks on three 7-point trait dimensions, each of which had a positive and a negative pole: (1) violent/not-violent, (2) unintelligent/intelligent, and (3) lazy/hard-working. Each item was coded such that higher scores indicated more positive ratings. A difference score for each trait was then created by subtracting ratings of blacks from ratings of whites. These three difference scores were then averaged to form a single index ( ) . a p .74

general egalitarianism
This variable, which measured respondents' support for egalitarian social values and arrangements, was indexed in both data sets using the six items from the NES egalitarianism scale developed by Feldman (1988). Higher scores indicated a greater level of egalitarianism ( in 1986; a p .74 a p in 1992). .75

opposition to affirmative action
In each data set, opposition to affirmative action programs was indexed using two items asking respondents to indicate their feelings about (1) preferential hiring and promotion for blacks and (2) the use of racial quotas in college admissions. Both items were coded such that higher scores indicated greater opposition to affirmative action ( in 1986; in 1992). a p .75 a p .69 demographics Finally, in each data set, five demographic variables typically related to white racial attitudes were also considered (Sears et al. 1997): age, gender, income, completion of a college degree, and a dummy variable indicating whether the respondent lived in a southern or non-southern state. In addition to providing a control for a variable known to affect racial attitudes, the inclusion of the college-degree variable in the models also allowed us to distinguish the "constraining" effects of sophistication from the somewhat-related effects of schooling (Bishop 1976;Converse 1980;Fiske et al. 1990). Since the completion of a higher degree is the educational experience most commonly linked to liberalized racial attitudes (see Steinberg and Selz-nick 1969), we decided that it would be the most appropriate way of controlling for said effects.

Analysis and Results
prejudice, race-neutral values, and affirmative action: effects of political sophistication As noted earlier, a number of researchers have argued that the policy reasoning of whites who are better able to understand the ideas that animate political life-as well as the essentially egalitarian norms of American political culture-will be driven more heavily by abstract ideological considerations, thereby lessening the effects of racism on opinions about policies like affirmative action. In contrast, others have suggested that whites with a better understanding of politics will be more likely to see how redistributive racial policies work against the material and symbolic interests of their group, thereby strengthening rather than weakening the relationship between racism and attitudes toward affirmative action. In order to examine these predictions, we regressed opposition to affirmative action on racial prejudice and several race-neutral political values among subsamples of whites who were low and high in political sophistication in each data set. In the 1986 NES, classical racism was used as the measure of racial prejudice, while conservatism, beliefs about government, and individualism served as the race-neutral controls. In the 1992 NES, the evaluative racial superiority measure was used, with conservatism and beliefs about government serving as the controls. In order to better distinguish whites with low and high levels of political sophistication, we trichotomized the sample on the basis of their political knowledge scores and used only the respondents in the bottom (low sophistication; in 1986; in 1992) and N p 316 N p 687 top (high sophistication; in 1986; in 1992) thirds of the N p 328 N p 608 sophistication distribution. In both data sets, opposition to affirmative action was then regressed separately on the attitude measures and demographics in each sophistication group. In order to determine whether any differences in the predictive power of our independent variables across these two levels of sophistication represented a reliable trend in each of the overall samples and to ensure that their scope was not restricted to any particular segment of the sophistication continuum, the b-coefficients for the interactions between each attitude and sophistication were examined in additional analyses, which used all of the respondents from each data set (and not merely those from the bottom and top thirds of the sophistication distribution).
For the 1986 NES, the results of this analysis are shown in the top panel of table 1. Looking first at the coefficients for political conservatism, we find a result consistent with the notion that sophistication may facilitate the use Note.-Multivariate entries for low-and high-sophistication groups are unstandardized regression coefficients, taken from models containing all attitudes and age, gender, income, completion of a college degree, and region. Interaction statistics are from models containing all three interactions, the main effects, and the demographics, using the full sample of whites. ϩ p ! .10, two-tailed. * p !. 05, two-tailed. ** p ! .01, two-tailed. *** p ! .001, two-tailed.
of race-neutral concepts in policy reasoning: even after the effects of racism, beliefs about government, and individualism were considered, the relationship between conservatism and opposition to affirmative action was positive and highly significant among whites with a better knowledge of politics, but nonsignificant among whites with a less sophisticated knowledge of politics. Examining the tests shown in the last column of table 1, we find a significant interaction between conservatism and sophistication ( , ), sug-b p .69 p ! .01 gesting that the difference between the coefficients in the low-and highsophistication groups represents a significant trend in the overall sample. In contrast, the results for our other race-neutral predictors were less supportive of this basic prediction. Controlling for the effects of conservatism and racism, the relationship between individualism and opposition to affirmative action was only slightly stronger among experts than it was among the less sophisticated; the associated interaction between individualism and sophistication in the full sample was nonsignificant as well ( , ). Moreover, b p Ϫ.09 p 1 .10 net of the other variables' effects, beliefs about government were not related to affirmative action attitudes in either group (both ; the interaction p's 1 .10 was also nonsignificant, ). p 1 .10

Sophistication and Racial Policy Attitudes 157
The results involving classical racism painted a somewhat different picture. Contrary to the notion that a sophisticated understanding of politics should attenuate the impact of prejudice, higher levels of classical racism were associated with greater opposition to affirmative action among whites with a strong understanding of politics ( , ), even after the effects of b p .27 p ! .01 the two race-neutral predictors were accounted for. In fact, opposition to affirmative action among our less sophisticated respondents, rather than being more heavily driven by prejudicial considerations, appeared to be associated with lower racism scores ( , ), suggesting that whites with b p Ϫ.25 p ! .001 a poor understanding of politics are not even able to connect a belief in the inferiority of blacks with opposition to affirmative action in an "appropriate," instrumentally rational fashion. Moreover, in contrast to what we found with regard to most of the race-neutral measures, the interaction between racism and sophistication in the full sample was highly significant ( , b p .81 p ! ), suggesting that the difference between these two coefficients represents .001 a reliable trend. 4 The analyses using the 1992 NES produced similar results, which are shown in the bottom panel of table 1. We again found that the effects of conservatism and beliefs about government were stronger at high levels of sophistication than at low levels of sophistication; however, only conservatism interacted significantly with sophistication in the overall sample ( ). However, p ! .001 the results for the evaluative racial superiority measure were largely consistent with the argument that sophistication may actually boost the effects of racism. Even after the effects of conservatism and beliefs about government were accounted for, the evaluative racial superiority measure was more strongly and positively related to affirmative action opposition in the high-sophistication group ( , ) than in the low-sophistication group ( b p .29 p ! .001 b p , ). As the tests in the final column of table 1 show, this variable .17 p ! .10 4. This is not the first time this somewhat counterintuitive negative correlation between classical racism and opposition to government help for blacks has been observed. In a sample of Bay Area respondents, it was found that those who felt blacks had been made inferior by God were also relatively supportive of government policy to aid blacks (see Apostle et al. 1983). This counterintuitive finding can be understood in attributional terms. That is, if blacks have been created inferior by God, they bear no direct responsibility for their plight and are therefore to be pitied rather than punished (Weiner 1986). An alternate possibility is that this result is due to acquiescence bias. Since both the racism and affirmative action scales consist of agree-disagree and favor-oppose items worded in the same direction, there may have been a tendency to agree with the racist sentiments in the racism items and favor the affirmative action programs, especially among nonsophisticates. To deal with this, we created an acquiescence scale by looking at responses to the eight "traditional values" items (a balanced set of items, with four coded in the "agree" direction and four coded in the "disagree" direction; 1986 NES items v525-v532) and counting the number of items to which a respondent gave "agree strongly" or "agree somewhat" responses (Sears, Henry, and Kosterman 2000). Corrected racism and affirmative action scales were then created by regressing the items in each scale on acquiescence, obtaining residual scores, and then recalculating the scale values for each measure using these residuals. When the analysis was rerun using these corrected scales, the results were identical, suggesting that our findings were not methodological artifacts.
interacted significantly with sophistication in the overall sample (both p's ! ), suggesting a reliable trend toward a stronger relationship between racial .01 prejudice and affirmative action opposition among respondents who best understand politics. 5 On the whole, then, these results provide little support for the hypothesis that the policy attitudes of whites with higher levels of political understanding would be less related to prejudice. On the one hand, the policy preferences of whites with high levels of political sophistication were indeed more strongly related to conservatism, although this was not consistently the case with beliefs about government and individualism. On the other hand, contrary to this basic hypothesis, opposition to affirmative action was also more heavily driven by attitudes of racial superiority among experts independent of the effects of the race-neutral predictors. 6 5. While these results are interesting in their own regard, they leave open the possibility that the interactions between racism and sophistication in each data set may be driven mainly by a relationship between sophistication and decreased opposition to affirmative action among those low in racism, without being accompanied by an additional relationship between sophistication and increased opposition to affirmative action among those high in racism. In order to check on this possibility, we regressed opposition to affirmative action on sophistication in the least and most racist thirds of each sample. Among whites low in racism ( in 1986; in N p 216 N p 340 1992), we found that sophistication was associated with decreased opposition to affirmative action in the 1986 NES ( , ) and completely unrelated to affirmative action attitudes in b p Ϫ.26 p ! .01 the 1992 NES ( , ). However, among whites high in racism ( in 1986; b p .06 p 1 .10 N p 320 in 1992), we found that sophistication was associated with increased opposition to N p 368 affirmative action in both datasets ( , in 1986; , in 1992). As such, b p .19 p ! .01 b p .14 p ! .01 it would seem that a sophisticated understanding of politics strengthens the relationship between racism and opposition to affirmative action not merely by reducing opposition among those who are not racist but also by boosting opposition among those who are. 6. Despite its insistence that racism and variables like conservatism and opposition to affirmative action should be intrinsically related to one another, the group dominance model also suggests that the nature of this relationship will depend on which aspects of the general construct of "racism" one focuses on (see Sidanius et al. 1996). While many researchers have defined racial prejudice strictly in terms of anti-black affect (e.g., Sniderman and Piazza 1993), it is clear that prejudice has an additional and perhaps more important component, i.e., the idea that subordinate groups are inherently inferior to dominant groups (i.e., "classical racism"; see Sidanius et al. 1996). Although these two aspects of prejudice are clearly related, they are far from identical. In this vein, several theorists (e.g., Jackman 1994; van den Berghe 1967) have noted that a belief in the inferiority of a given outgroup-and desires for group dominance-need not be accompanied by negative affect toward that group (as in the antebellum South; see van den Berghe 1967; see also Sidanius and Pratto 1999). However, despite historical variance in the role of affect in prejudice, the idea of black inferiority has remained a constant element in white racial attitudes for most of American history. Consistent with this reasoning, we have relied on measures rooted in superiority/inferiority beliefs. Nevertheless, one might argue that the use of measures like these prevents our analysis from truly addressing existing politics-centered work on the role of political understanding in racial attitudes, which implies that a healthy awareness of abstract political ideas and mainstream political norms should mainly reduce the influence of affect on policy attitudes (see Sniderman, Brody, and Tetlock 1991). However, when we substituted antiblack affect (item v149 in the 1986 NES, item v5323 in the 1992 NES) for the prejudice measure in each data set, we still did not find that the net effect of anti-black affect was smaller in the high-sophistication groups. In the 1986 NES, its effect was the same in both groups ( , b p .20 both ), while its effect actually increased slightly in the 1992 NES (i.e., , p's ! .05 b p .21 p ! , in the low-sophistication group, vs. , , in the high-sophistication group). .01 b p .30 p ! .001 Moreover, when anti-black affect was added to the full 1986 and 1992 models shown in table the relationship between political conservatism and racial prejudice Consistent with the argument that most Americans have internalized the message of the civil rights movement and accepted racial equality as one of the assumptions of the American Creed, research suggesting that education and political sophistication may weaken the effects of prejudice also implies that political conservatism and prejudice will be linked only among whites who do not truly understand the color-blind nature of conservative values, that is, political novices (see ; see also McClosky and Zaller 1984). In contrast, consistent with the argument that racism and conservatism may share a common root-namely, a desire for unequal relationships among groups-others have suggested that they should be more strongly related among whites who truly understand the implications of various political ideas (e.g., Sidanius et al. 2000).
In order to examine these predictions, we regressed conservatism on the racial prejudice measures from 1986 and 1992 and the demographics among whites who were low and high in political sophistication in each data set. Once again, we used only respondents from the bottom and top thirds of the knowledge continuum in each survey, and full-sample interaction tests similar to those reported above were used to determine whether any differences in the predictive power of each racism measure across these two groups were reliable.
For the 1986 NES, the results of this analysis are shown in the top row of table 2. Here, conservatism was regressed on classical racism at high and low levels of sophistication. Looking first at the coefficients for this relationship, we once again obtain a result at odds with the notion that the beliefs of sophisticated whites should be less contaminated by prejudice. While this relationship was significant and positive in the high-sophistication group, it failed to reach significance in the low-sophistication group. The test on the interaction between racism and sophistication in the full sample produced a significant result ( , ), suggesting a reliable trend. A parallel b p .24 p ! .01 analysis using the 1992 NES produced similar results, shown in the bottom row of table 2. Looking at these coefficients, we can see that the relationship between evaluative racial superiority and conservatism was significant and positive in the high-sophistication group, but nonsignificant in the low-sophistication group. Moreover, the significance of this result is reinforced by the presence of a reliable interaction between evaluative superiority and sophistication in the full sample ( , ). b p .63 p ! .001 Thus, our analyses of the relationship between racism and conservatism among groups of whites differing in political sophistication were not sup- Note.-Multivariate entries for low-and high-sophistication groups are unstandardized regression coefficients, taken from models containing racial attitudes and age, gender, income, completion of a college degree, and region. Interaction statistics are taken from models using the full sample of whites from each data set. ϩ p ! .10, two-tailed. * p ! .05, two-tailed. ** p ! .01, two-tailed. *** p ! .001, two-tailed.
portive of the idea that a better understanding of political ideas and the norms of mainstream political culture would weaken the relationship between prejudice and ideology. Contrary to this notion, conservatism was more strongly associated with racism among sophisticated whites. 7

egalitarianism and opposition to affirmative action
Analyses suggesting a genuinely reduced role for prejudice in whites' racial policy attitudes have also implied that a distaste for equality is no longer an important antecedent of opposition to such policies and that a commitment to certain types of equality may in fact drive whites' opposition to affirmative action, particularly among those with a sophisticated understanding of politics. That is, whites who oppose affirmative action may do so out of a belief that the policy is itself discriminatory and contrary to the formally egalitarian norms of the American Creed, providing certain groups with an unfair ad-7. Again, in order to test our hypotheses using the feeling-thermometer measures emphasized by the politics-centered approach, we repeated the analyses in this section using anti-black affect instead of the prejudice measures. However, in both data sets, anti-black affect was not related to conservatism in either sophistication group (all ). Moreover, when anti-black affect p's 1 .10 was added to the analyses reported in table 2 as an additional control, it did not change the results involving the prejudice measures: as before, prejudice was unrelated to conservatism among those low in sophistication ( in 1986 ; in 1992; both ) but b p .03 b p Ϫ.07 p's 1 .10 positively associated with conservatism among sophisticates ( , in 1986; b p .17 p ! .05 b p , in 1992). Thus, these results reinforce the notion that political attitudes will be only .31 p ! .01 weakly related to racism when the latter is defined strictly in terms of racial animus or antiblack affect, and they also suggest that the effect of racism on ideology is relatively independent of the effect of group affect (see Sidanius et al. 1996). vantage in hiring, school admissions, and the like (Sniderman and Carmines 1997;Thernstrom and Thernstrom 1997; see also Connerly 1996). As Sniderman and Piazza (1993) have argued, "Though there are elements of the Creed that can be deployed in favor of affirmative action, the fundamental ideas of fairness and equal treatment, for most ordinary citizens, thrust in exactly the opposite direction. What gives the race-conscious agenda its distinctive character, what makes the agenda open to challenge morally, is that the principle of preferential treatment runs against the Creed" (p. 177). Thus, this approach suggests that opposition to affirmative action may be positively related to egalitarian sentiment, and-at any rate-that it should not be significantly driven by antiegalitarianism in the present social and historical context, at least among those who clearly understand the norms of American political culture. In contrast, other researchers maintain that whites' opposition to affirmative action still stems in part from a desire to protect their group's privileged position in the racial hierarchy . As such, one should find that antiegalitarian whites are most opposed to affirmative action. Moreover, this should particularly be the case among whites who have a better understanding of political ideas.
In other words, there appears to be a basic disagreement about the role of egalitarian values in the genesis of whites' racial policy attitudes, particularly with regard to sophisticates. Nevertheless, the arguments at the heart of this disagreement have not been subjected to a great deal of empirical scrutiny (but see Sidanius et al. 2000). In order to shed some light on this problem, we first of all regressed opposition to affirmative action on general egalitarianism and the demographic controls among all of the whites in each data set. Since the items in this measure cover several aspects of people's beliefs about equality-including support for equal opportunity, fears that outcomes have become too unequal in American society, and the general importance of equality as a social value (see Feldman 1988)-this analysis allowed us to get a broad look at the relationship between egalitarianism and racial policy attitudes. Contrary to the suggestion that opposition to affirmative action may have egalitarian roots, we found that a preference for equality was negatively related to affirmative action opposition in 1986 ( , ) and b p Ϫ.32 p ! .001 1992 ( , ), net of the effects of the demographic controls. b p Ϫ.38 p ! .001 Despite the strength of these results, a possible objection to this analysis involves our broad operational definition of egalitarianism. In discussion of the role played by egalitarian beliefs in whites' opposition to affirmative action, a number of theorists have been careful to point out that the equality whites are most likely to have in mind when making such judgments is equality of opportunity-that is, equal treatment and an equal chance to succeed, rather than equal "results" in the game of life (see Sniderman and Carmines 1997;see also Pole 1993;Rae 1981). Since our measure of egalitarianism contained items measuring support for both types of equality, the analysis described above may have provided a less than optimal test of the question at hand. Therefore, we repeated the analysis, this time using a modified egalitarianism scale based only on the three items from the NES egalitarianism that have been specifically identified as measures of support for equality of opportunity (i.e., the so-called more equal treatment items; see Sears, Henry, and Kosterman 2000). 8 However, like general egalitarianism, endorsement of these equal opportunity items was also negatively related to affirmative action opposition in both the 1986 NES ( , ) and the 1992 NES b p Ϫ.24 p ! .001 While these results are instructive, they do not get at a key issue-the suggestion that opposition to affirmative action should be a matter of outright antiegalitarianism primarily among those who lack the political knowledge necessary for a genuine understanding of the values at the heart of the American Creed. In order to examine this possibility, we repeated the above analyses separately for respondents at low and high levels of political sophistication in each data set. As before, we restricted the analysis to respondents in the bottom and top thirds of the political knowledge distribution for illustrative purposes. Full-sample interaction tests similar to those reported earlier were used to determine whether differences in the relationships between our egalitarianism measures and affirmative action opposition across these two groups represented a reliable trend. The results of these analyses are displayed in table 3. Looking first at the results for 1986 NES, shown in the top panel of this table, we can see that affirmative action attitudes were essentially unrelated to general egalitarianism among whites with a poor understanding of politics. However, contrary to the hypothesis that sophistication may dampen the effects of antiegalitarian sentiment, egalitarianism had a strongly negative relationship with opposition to affirmative action among sophisticated whites. Moreover, this difference was accompanied by a highly significant interaction between egalitarianism and sophistication in the full sample ( , ). Similarly, support for b p Ϫ.77 p ! .001 equal opportunity was unrelated to affirmative action opposition among those low in sophistication, but negatively related to affirmative action opposition among respondents high in sophistication-a result that was again reinforced by the presence of a statistically reliable interaction between equal opportunity and sophistication in the overall sample ( , ). The b p Ϫ.37 p ! .05 analyses using the 1992 NES produced an analogous pattern of results, which 8. In 1986, these items were v364, v366, and v369; in 1992, they were v6024, v6028, and v6029. The reliabilities for these equality of opportunity scales were (in the 1986 NES) a p .61 and (in the 1992 NES). Note that these items are agree-disagree items, all worded in a p .65 the same direction. Since the affirmative action items also had a favor-oppose format, this again raises the possibility of acquiescence bias. In order to deal with this, we corrected each scale for acquiescence in the same manner as before (see n. 4 above). In the 1986 NES, this was done using the same acquiescence scale described earlier. In the 1992 NES, it was done using a similar scale based on four of the "traditional values" items (two were coded in the "agree" direction, and two in the "disagree" direction; see 1992 NES items v6115, v6116, v6118, and v6119). Analyses using these corrected scales produced results identical to those reported in this section. Note.-Entries for low-and high-sophistication groups in the "With Demographics" columns are unstandardized regression coefficients, taken from models containing the indicated attitude and age, gender, income, completion of a college degree, and region; the effects of each equality index were tested in separate models. Interaction statistics are from models also containing the demographics and the main effects for the attitude and political sophistication, using the full sample of whites from each data set. Each row represents a separate model. ϩ p ! .10, two-tailed. * p ! .05, two-tailed. ** p ! .01, two-tailed. *** p ! .001, two-tailed.
are shown in the bottom panel of table 3. While the relationship between general egalitarianism and affirmative action opposition was significant in both sophistication groups, it was much larger among those with high sophistication, a difference that was again associated with a highly significant interaction between egalitarianism and sophistication in the overall sample ( , ). Similarly, the relationship between support for equal b p Ϫ.74 p ! .001 opportunity and opposition to affirmative action was significant in both sophistication groups, but somewhat larger in the high-sophistication group. The interaction between equal opportunity and sophistication was significant in the overall sample as well ( , ), suggesting that the dif-b p Ϫ.23 p ! .05 ference between the above coefficients was reliable.
how consensual a norm is equality of opportunity?
In the next set of analyses, we examined the assumption that support for formal equality-that is, equality of opportunity-is a fairly consensual norm in American political culture. In this vein, work suggesting a reduced role for prejudice in contemporary white racial policy attitudes implies that one of the long-term effects of the civil rights movement was to bring about a general consensus in favor of equal opportunity. Indeed, when we looked at responses to the three items in the equal opportunity scales from each data set, we found that most people gave responses in favor of equal opportunity: across the three items in each data set, an average of 64 percent of the respondents in the 1986 NES and 73 percent of the respondents in the 1992 NES either "agreed strongly" or "agreed somewhat" with the items' pro-equal-opportunity sentiments. However, it may be that this general pattern of concern for equal opportunity, while stronger than in the past, is still not evenly distributed across the ideological spectrum, particularly among those who understand the implications of their beliefs for the maintenance of group hierarchies.
In order to examine this possibility, we first of all regressed the equal opportunity measure from the previous set of analyses on conservatism and the demographics in each data set. 9 All of the white respondents from each survey were used in these regressions. The results of this analysis showed that conservatism was negatively related to support for equal opportunity in both data sets ( in 1986; in 1992; both ), net b p Ϫ.32 b p Ϫ.35 p's ! .001 of the demographics. However, since we have already shown that conservatism is linked to various forms of racial prejudice, it is possible that a portion of the negative relationship between conservatism and concern for equal opportunity may be due to a shared relationship with racism. Thus, we repeated the above analysis, this time including the prejudice measures from each data 9. The analyses described in this section were also run a second time using the acquiescencecorrected measures of equal opportunity (in 1986 and 1992) and racism (in 1986) discussed in nn. 3 and 7 above. These analyses again produced identical results, suggesting that our findings were not methodological artifacts. Note.-Multivariate entries for low-and high-sophistication groups are unstandardized regression coefficients, taken from models containing all attitudes and age, gender, income, completion of a college degree, and region. Interaction statistics are from models containing the demographics and the main effects for all attitudes and political sophistication, using the full sample of whites from each data set. ϩ p ! .10, two-tailed. * p ! .05, two-tailed. ** p ! .01, two-tailed. *** p ! .001, two-tailed.
set as controls. The results of this analysis showed that conservatism was still negatively related to support for equal opportunity in both data sets (b p in 1986; in 1992; both ), even after the effects of Ϫ.32 b p Ϫ.34 p's ! .001 classical racism (in the 1986 NES) and evaluative racial superiority (in the 1992 NES) were taken into account. As such, these results suggest that concern for equal opportunity is not evenly distributed across the political spectrum, despite perhaps being stronger and more well diffused than in the past.
However, the key issue of whether this relationship differs across sophistication levels still remains. In order to address this question, we regressed the equal opportunity measure on conservatism at low and high levels of political sophistication in each data set. As before, only respondents in the bottom and top thirds of the sophistication distribution were included in the analyses for each data set, although full-sample interaction tests were again used to examine any differences that emerged. For the 1986 NES, the results of this analysis are shown in the top panel of table 4. As a first step, we regressed equality of opportunity on conservatism alone in the low-and highsophistication groups (model 1). Here, we find that conservatism and attitudes toward equal opportunity were essentially unrelated among whites with low levels of political sophistication, but strongly negatively related among those high in sophistication. Consistent with this pattern, the interaction between conservatism and sophistication in the overall sample was significant (b p , ). Moreover, when we repeated this analysis with racism in the Ϫ.60 p ! .001 model as a control, this basic result remained the same (see model 2). The analyses using the 1992 NES produced similar results, which are shown in the bottom panel of table 4. When equal opportunity was regressed on conservatism and the demographics alone (model 1), we found that conservatism and concern for equal opportunity were negatively related in both the lowand high-sophistication groups, although the relationship was twice as strong among those high in sophistication. Consistent with this general pattern, the interaction between conservatism and sophistication in the full sample of whites was significant ( , ). In order to ensure that this re-b p Ϫ.52 p ! .001 lationship was not due to the confounding effects of racism, we once again repeated this analysis, this time controlling for the effects of evaluative racial superiority. As before, this did not change the overall pattern of findings (see model 2). 10 On the whole, then, these results suggest that concern for equal opportunity is not necessarily a politically consensual norm. Rather than being orthogonal to beliefs about equal opportunity, conservatism was associated with weaker support for equal opportunity norms in our two samples. Moreover, instead of diminishing this negative relationship, political sophistication appeared to 10. Breaking these interactions down slightly differently, we also find that sophistication has a different relationship with concern for formal equality among liberals and conservatives (McClosky and Zaller 1984;Zaller 1992). In the most liberal third of each sample ( in N p 269 1986; in 1992), we find that sophistication is associated with increased concern for N p 349 equal opportunity ( in 1986; in 1992; both ). However, in the most b p .18 b p .10 p's ! .01 conservative third of each sample ( in 1986; in 1992), we find that sophis-N p 305 N p 329 tication is associated with decreased concern for equal opportunity ( , in 1986; b p Ϫ.13 p ! .05 , in 1992). Moreover, net of the demographics in each full sample, we found b p Ϫ.20 p ! .001 virtually no relationship between sophistication and concern for equal opportunity (i.e., b p , in 1986; , in 1992). Similarly, we actually found more variance Ϫ.03 p 1 .10 b p .05 p p .08 in equal opportunity scores among the sophisticated than among the unsophisticated (i.e., .93 vs. .72 in 1986; .79 vs. .62 in 1992); both of these differences were significant, as indicated by Bartlett's test for homogeneity of variances (both ). If there were general consensus in p's ! .05 the political culture about equal opportunity, we would expect both stronger relationships between sophistication and equal opportunity and less variance in the high-sophistication group, indicating that individuals who were generally more aware of elite norms were also generally more supportive of equal opportunity and less likely to differ among themselves in their opinions about it. Instead, what we appear to have is a situation where sophistication polarizes attitudes toward equal opportunity on the basis of respondents' prior beliefs, resulting in greater variance in equal opportunity beliefs among sophisticates, consistent with the pattern one would expect given elite disagreement on an issue (see McClosky and Zaller 1984). increase its magnitude, suggesting that conservatives who best understand their beliefs-and the norms of American political culture-are even less likely to be concerned about equality of opportunity. In fact, the only respondents for whom ideology and concern for formal equality were orthogonal were our novices, who should be less likely than other individuals to truly understand the implications of the beliefs they espouse. So, while we have no reason to believe that equal opportunity is not endorsed by most Americans, the present findings lead us to also conclude that it is not uniformly endorsed across the political spectrum.

knowledge and education as moderators of key relationships
Finally, one of the key implications of the perspective advanced here is that knowledge should be a better indicator of political understanding than education. If this is the case, then we should also find that it has stronger effects on the magnitude of the key relationships examined in the earlier sections. In order to confirm this, we looked at the degree to which each variable moderated the relationships indicated in table 5. Two new sets of interaction analyses, parallel to those reported in tables 1-4, were conducted: one that tested the interactions between each predictor and years of education alone, and another that simultaneously tested the interactions between each predictor and education and the interactions between each predictor and knowledge. For comparative purposes, the interaction results for knowledge alone, taken from the tables 1-4, are reported in the knowledge columns labeled "Tested Separately." The results of these analyses are summarized by the totals at the bottom of each column. Looking first at the analyses in which the moderating effects of knowledge and education were examined separately, we can see that knowledge moderated a greater number of relationships in each data set (i.e., 6 vs. 3, with one additional marginal interaction with education in 1986; 6 vs. 4 in 1992). This pattern becomes even clearer when we look at the results of the analyses that examined the effects of both moderators simultaneously. In 1986, knowledge moderated five of the relationships, while education moderated only one. Similarly, in 1992, knowledge moderated four of the key relationships, while education moderated only two of them. Moreover, in 1986, the results of the simultaneous tests also indicated that there were five cases in which knowledge alone moderated a key relationship, while there was only one case in which education alone moderated a relationship. Analogously, in 1992, there were three cases in which knowledge alone fully or marginally moderated key relationships, but none in which education alone had a moderating effect. Thus, these results provide suggestive evidence for the idea that knowledge may be a better indicator of political understanding-and its effects on relationships between attitudes-than education. Note.
-Entries indicate the significance of the interaction between the variable in each row and the moderator (knowledge or education) shown at the top of each column. The "Tested Separately" entries for the knowledge interactions are taken from the tests shown in tables 1-4. All analyses were multivariate. ϩ p ! .10, two-tailed. * p ! .05, two-tailed. ** p ! .01, two-tailed. *** p ! .001, two-tailed.

Summary and Discussion
As we noted earlier, the hypothesis that a better understanding of abstract political concepts and the basic norms of American political culture should weaken the relationship between racism and policy attitudes and produce a more consensual pattern of support for equal opportunity among whites has received an increasing amount of empirical attention (e.g., Sidanius et al. 1996Sidanius et al. , 2000; see also . Nevertheless, the results of these studies are subject to criticism with regard to the manner in which they have operationalized individuals' understanding of political concepts and mainstream political norms. Most of these studies have relied on education as their primary index of political understanding (see Sidanius et al. 1996Sidanius et al. , 2000; for similar approaches, see Jackman and Muha 1984). While education certainly has implications for one's level of political sophistication, recent work suggests that it may not adequately capture the construct of political understanding, at either a conceptual or an empirical level (Fiske et al. 1990;Zaller 1990). As such, previous studies may not have provided an adequate examination of the hypothesis that an awareness of political ideas and mainstream political norms may attenuate the impact of prejudice and antiegalitarianism. Therefore, in an effort to address this problem, we have defined political sophistication in terms of actual political knowledge in the present study. Using this index in two different national probability samples, we examined the relationships between affirmative action attitudes, prejudice, race-neutral political considerations, and beliefs about equality among white respondents with different levels of political sophistication.
In both data sets, our results provided little or no support for the notion that a better understanding of abstract political ideas and the norms of American political culture would attenuate the significance of racist and antiegalitarian motives. First of all, beliefs about racial superiority and inferiority were more strongly associated with affirmative action attitudes among the sophisticated than they were among relatively unsophisticated whites. Furthermore, rather than being less strongly related to superiority beliefs among sophisticates, political conservatism was, in fact, more strongly related to racism among these individuals. Perhaps more interestingly, we found little support for the notion that opposition to affirmative action may be driven by egalitarianism, particularly among sophisticated respondents. Not only was egalitarianism negatively (rather than positively) related to affirmative action opposition, but the magnitude of this relationship actually increased with political sophistication. Moreover, this general pattern of results held up even when we restricted our definition of egalitarianism to equality of opportunity only. Thus, while many prominent opponents of affirmative action have argued that opposition to this policy is rooted in an endorsement of Martin Luther King's original dream of equal opportunity-as well as the formally egalitarian values at the heart of American political culture (e.g., Connerly 1996)-a growing body of evidence suggests that just the opposite holds in mass publics. Rather than being embedded in support for egalitarian values, opposition to affirmative action shows every appearance of being driven by antiegalitarian values, particularly among those who should best understand the principles of the American Creed.
Moreover, our results also imply that equality of opportunity is less of a consensual value in American society than many have suggested, especially among the politically knowledgeable. That is, in both samples, political conservatives were generally less supportive of equality of opportunity. This was particularly the case among knowledgable conservatives, who should be most attuned to the values at the heart of conservative thinking. Thus, while most Americans do profess a relatively strong level of support for this basic value (see McClosky and Zaller 1984; see also Schuman et al. 1997), they do not do so evenly across the ideological spectrum. More broadly, these findings-along with our primary analysis of the antecedents of racial policy attitudes-also imply that the content of the American Creed may be more complex and contradictory than many students of race relations have been willing to admit. Although the notion of a relatively unitary American political culture centered on the principle of formal equality has held a great deal of sway in studies of racial politics (e.g., Myrdal 1944; see Dawson 2000 for a discussion), recent historical work suggests a far less homogenous Creed, one in which unambiguously racist and antiegalitarian impulses are often found right alongside egalitarian ones (Mills 1997;Smith 1997). Our findings not only reinforce this claim at an empirical level, but they also suggest that this ethos of contradiction may have survived a period in which racial attitudes underwent an extensive process of liberalization. That is, despite changes in public opinion that have greatly reduced overt prejudice and broadened support for the value of equal opportunity, we still find a considerable level of political disagreement with regard to the importance of this value, as well as a continued connection between racism, policy attitudes, and people's ideological convictions. The fact that these trends were even more marked among those most aware of the content of American political culture-that is, political sophisticates-would seem to make the complex nature of the American Creed even more apparent. 11 Finally, at a more technical level, our comparative analysis of the moderating effects of political sophistication and education lent support to the idea that the latter variable may not be the most reliable indicator of the degree to which survey respondents understand abstract political ideas and the norms of mainstream American political culture. More precisely, our results indicated that sophistication was generally more likely to moderate the key relationships 11. One objection to our findings involves the possibility that sophistication-based differences in the strength of the relationships between key variables may be an artifact of different levels of measurement error and/or variance across the two groups (Judd, Krosnick, and Milburn 1981). In order to deal with this problem, we obtained corrected estimates of these relationships using Herting's (1985) LISREL-based method for disattenuating regression coefficients for unreliability. In these analyses, the regressions reported in tables 1-4 were repeated at each sophistication level using multigroup LISREL models in which each variable was specified as a latent factor indicated by one measured variable, i.e., respondents' scores on the scale for that variable. To correct the actual estimates, the single factor loading for each variable was set to the square root of its reliability in each sophistication group, and the error variance for the single indicator was set to the square root of , where R is the reliability of the variable in that group. To test 2 1 Ϫ R the differences between the corrected coefficients for each predictor across groups, they were constrained to equality across the groups in separate model runs; a reliable difference was indicated when this constraint produced a significant decline in model fit (as indicated by a chisquare difference test, which also has the advantage of being a more conservative test than the t-test used to assess interactions in regression; see Jöreskog and Sörbom 1993). Unstandardized regression coefficients, which are robust with regard to between-group variance differences (see Pedhazur 1982, pp. 30-32, for a demonstration), were used in all analyses as an additional safeguard. The results of these analyses were identical to those reported above, with one exception: in the 1992 NES, the net relationship between evaluative racial superiority and affirmative action opposition was significant in both groups (i.e, , in the low group, , in the high b p .27 b p .28 group; both ). The two coefficients did not differ significantly ( ). While this p's ! .001 p 1 .10 result does not suggest that sophistication strengthened the relationship between prejudice and affirmative action attitudes, it still runs counter to the prediction that racism should have less of an impact among whites with better political understanding. examined in this study. This pattern was particularly apparent when we directly compared the moderating effects of each variable in a series of simultaneous interaction models. Thus, while we do not discount the important role of education in the shaping of racial attitudes-particularly with regard to its tendency to reduce overall levels of prejudice (see Steinberg and Selznick 1969; see also Federico and Sidanius 2000)-our findings suggest that researchers interested in the role of abstract political understanding and awareness in the organization of racial attitudes into clusters of related beliefs may want to rely more heavily on direct measures of understanding, such as the knowledge indices used here.
In conclusion, while it is clear that racial policy attitudes are not merely a function of racist and antiegalitarian motives-and that the purely political considerations that animate mainstream discourse are as relevant to debates about racial policy as they are to disputes in other policy areas (see Sniderman et al. 2000 for a discussion)-there may now be reason to regard the hypothesis that a better understanding of political ideas and the norms of the broader political culture should attenuate the impact of prejudicial factors with a certain amount of skepticism, given the number of disconfirmations that have now emerged (see also Jackman and Muha 1984;Sidanius et al. 1996Sidanius et al. , 2000. Nevertheless, while there is thus a great deal of evidence calling this hypothesis into question with regard to both the mass public and various student populations (see Sidanius et al. 1996), it is still conceivable that the sophistication hypothesis may hold among members of certain clearly demarcated groups of political actors. Here, we are alluding to the population of "true political elites," including party leaders, academics, political consultants, journalists and newspaper editors, and various government officials (Converse 1964;McClosky and Zaller 1984). While the logic of existing theory and research in this area gives us no reason to expect this hypothesis to be more valid within this select population than in any other, it is also clear that principled opposition to affirmative action has resonated strongly with certain members of these groups, often inspiring eloquent and genuinely race-neutral arguments against the policy (see Connerly 1996;Glazer 1975). Moreover, students of public opinion have suggested that the beliefs of these elites may differ even from those of the most highly sophisticated portion of the mass public (see Converse 1980). As such, this issue cannot be considered settled until this most select of populations has been examined and our findings have been replicated. We hope that such research will be forthcoming. you rate whites in general on this scale?" (v6225), "Blacks?" (v6226). Peaceful: "Do people in these groups tend to be violent or do they tend to be peaceful?" "Where would you rate whites in general on this scale?" (v6229), "Blacks?" (v6230). 1986 NES and NES. "Our society should do whatever is necessary to make sure that everyone has an equal opportunity to succeed," "We have gone too far in pushing equal rights in this country," "One of the big problems in this country is that we don't give everyone an equal chance," "This country would be better off if we worried less about how equal people are," "It is not really that big of a problem if some people have more of a chance in life than others," and "If people were treated more equally in this country we would have many fewer problems" (v364-v369 in 1986; v6024-v6029 in 1992).

Opposition to Affirmative Action
1986 NES and 1992 NES. "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 of blacks is wrong because it discriminates against whites. What about your opinion-are you for or against preferential hiring and promotion of blacks?" and "Some people say that because of past discrimination it is sometimes necessary for colleges and universities to reserve openings for black students. Others oppose quotas because they say quotas discriminate against whites. What about your opinion-are you for or against quotas to admit black students?" (v476 and v478 in 1986; v5936 and v5948 in 1992).