The following is a guest post from Chris Elmendorf and Doug Spencer:
Are the Covered States “More Racist” than Other States?
Christopher S. Elmendorf
Douglas M. Spencer
During oral argument last week in Shelby County v. Holder, the constitutional challenge to Section 5 of the Voting Rights Act, Chief Justice Roberts asked, “[I]s it the government’s submission that the citizens in the South are more racist than citizens in the North?” Solicitor General Verrilli responded, “It is not, and I do not know the answer to that . . . .”
This post offers a preliminary answer to the Chief Justice’s question, using recent data. Our initial results suggest that the coverage formula of Section 5 does a remarkably good job of differentiating states according to the racial attitudes of their nonblack citizens.
There are essentially three schools of thought about how best to measure racial prejudice using survey questions. Some researchers favor explicit measures of prejudice (“old-fashioned racism” or stereotyping), based on agreement with statements like “blacks are less intelligent than whites” and “blacks are lazy.” Others favor symbolic measures of prejudice or “racial resentment,” based on questions about affirmative action and whether blacks have gotten “more than they deserve.” Still others favor measures of implicit or subconscious bias. For the results reported here we use explicit stereotyping, as it remains disputed whether racial resentment measures capture prejudice as opposed to conservatism, and it is uncertain whether implicit bias predicts political behavior.
We created a binary measure of stereotyping that roughly captures whether a person is more prejudiced toward blacks than is typical of nonblack Americans. Our data source is the 2008 National Annenberg Election Survey (NAES), which asked non-black respondents to rate their own racial group and blacks in terms of intelligence, trustworthiness, and work effort, on a scale of 0-100. On average respondents ranked their own group about 15 points above blacks on each trait. We coded respondents as holding “prejudiced” views with respect to blacks on a particular trait if the difference between their rating of their own racial group and their rating of blacks exceeded the national mean difference for the trait. To create an overall measure of prejudice for each respondent, we summed the number of traits on which the respondent was more prejudiced than the national mean. Finally, we converted this sum into a binary variable, coding as “prejudiced overall” those respondents who exceeded the national mean with respect to at least two of the three traits.
To be clear, a respondent whom we have coded as “not prejudiced overall” may well be quite prejudiced. But the Chief Justice’s question—whether “citizens in the South are more racist than citizens in the North”—is a question about relative prejudice, and this is what we are trying to capture.
We provide two estimates of the proportion of adult, nonblack residents in each state who are “prejudiced overall.” The first is based on simple disaggregation of the large NAES dataset (N=19,325). This method should work pretty well for the largest states but may yield unreliable estimates for smaller states, which contribute relatively few respondents to the NAES sample. For the second estimate we use multilevel regression with post-stratification (MRP), a recently developed statistical technique that has been shown to yield remarkably accurate estimates of state-level public opinion. We model prejudice as a function of individual-level covariates (sex, race, age, and education) and a set of state-level predictors (black population, percent of blacks in poverty, segregation, and income inequality).
Using either technique we find a strong positive correlation between Section 5 “covered status” and anti-black prejudice, but with MRP the correlation is truly stunning:
The MRP model suggests that the six fully covered states in the South are, by our measure, six of the seven most prejudiced in the nation. The two fully covered states that rank lower on the list, Arizona and Alaska, are presumably covered for reasons other than discrimination against blacks (anti-Latino discrimination in Arizona, and anti-Native discrimination in Alaska).
We wish to emphasize that these are preliminary results only. Though our findings are not entirely unexpected, other ways of aggregating the NAES prejudice questions, or of modeling responses, may yield different rankings of the states (to say nothing of other ways of measuring prejudice). We will present additional results at the Midwest Political Science Association conference in April.
Suffice it to say for now that the coverage formula seems defensible under the standard implicit in the Chief Justice’s questioning. Or, to borrow a metaphor from Judge Williams of the D.C. Circuit, Congress appears to have “hit the bull’s eye throwing a dart backwards over its shoulder.”
Elmendorf is Professor of Law at UC Davis. Spencer is a doctoral student in Jurisprudence and Social Policy at UC Berkeley. Elmendorf contributed to an amicus brief on behalf of the respondents in Shelby County v. Holder.
 Our overall measure of prejudice includes just those respondents who exceeded the national average by at least one standard deviation, or 14% of the sample.