Below, a piece by Chris Elmendorf on statistical tests for racism, and their impact on the Voting Rights Act. Originally published in JURIST (jurist.org), and available there at http://jurist.org/forum/2012/06/christopher-elmendorf-voting-future.php.
* * *
Enacted in 1965 and reauthorized several times since, the Voting Rights Act (VRA) has long been regarded as the capstone of our nation’s civil rights architecture. The VRA’s core provisions have, however, come under sharp criticism from commentators and jurists who consider them outmoded or even unconstitutional in a society that is far more racially tolerant than it was in 1965. Defenders of the VRA argue that it remains necessary because contemporary racial prejudices or the lingering effects of past discrimination represent ongoing barriers to the political aspirations of minority voters and candidates.
Beyond riling up the already converted, neither side has made much headway in this debate. Progress has been slowed by difficulties in measuring the severity and geography of contemporary racial prejudices, and the extent to which they shape voters’ decisions. Most scholars believe that survey-based measures of overt racial prejudice underestimate contemporary prejudice, because survey respondents do not want to be seen as racist or to see themselves as racist. Psychologists have developed alternative measures of “implicit bias” or “racial resentment” that do not require survey respondents to cop to being racist, but other scholars dispute the scientific [PDF] or normative [PDF] validity of these metrics. And whatever one makes of the metrics, they have not yet yielded a fine-grained picture of the geography of racial discrimination, which is necessary to resolve current controversies about the VRA.
My purpose in writing this commentary is not to describe the contours of seemingly intractable debate, but to argue that answers are finally at hand — thanks to path-breaking new research on the geography of discrimination. Seth Stephens-Davidowitz, a doctoral student in economics at Harvard, is deploying publicly available information to measure the frequency with which Google searches use the “n-word” in each of the nation’s 210 media markets. (It turns out that most searchers who used the n-word were looking for derogatory jokes about black people.) He shows [PDF] that in relatively prejudiced regions, then-candidate for president, Barack Obama, substantially underperformed relative to his expected vote share.
I argue here that Stephens-Davidowitz’s research is likely to be the nail in the coffin of Section 5 of the VRA, which requires jurisdictions in certain regions of the country to “pre-clear” changes to their election laws with the US Department of Justice or the District Court of the District of Columbia. But even as it hastens the demise of Section 5, Stephens-Davidowitz’s work should greatly strengthen Section 2 of the Act, which applies nationally and which prohibits election laws that “result” in minority voters having “less opportunity than other members of the electorate to participate in the political process and to elect representatives of their choice.”
The Coming Demise of Section 5
Section 5 of the VRA requires so-called “covered jurisdictions” — mostly states in the Deep South, plus a few states and localities elsewhere — to obtain the federal government’s approval before implementing changes to their election laws. The feds must deny approval if the change would make minority voters worse off, or if it was motivated by a discriminatory purpose. The burden of proof is on the jurisdiction seeking pre-clearance. Section 5 was originally a temporary measure but it has been extended several times, most recently in 2006 for another 25 years.
The constitutionality of the latest extension first reached the US Supreme Court in 2009. The Court used some interpretive acrobatics to avoid the constitutional question for the time being — while signaling that Section 5 may well be struck down as an improper exercise of congressional enforcement authority under the Fourteenth and Fifteenth Amendments unless Congress takes steps to improve the fit between Section 5’s reach and the occurrence of actual constitutional violations.
The problem as some judges and commentators see it is that, in extending Section 5, Congress did not even try to improve its constitutional fit. The “coverage formula” for determining which jurisdictions are subject to Section 5 continues to be based on election returns from the 1970s and state practices in the 1960s and 1970s. Nor did Congress update the “bailout provisions,” which determine whether a jurisdiction may escape from Section 5’s coverage.
Defenders of Section 5 argue that the coverage formula is constitutionally adequate because it captures most of the former Jim Crow South, where intentional racial discrimination in the electoral process is likely to be worse than in other states. They also point out that adjudicated Section 2 violations have disproportionately occurred in covered jurisdictions, as have judicial findings concerning racial appeals in campaigns and related indicia of intentional discrimination by the electorate as a whole. But there are deep and, among empiricists, widely appreciated problems with drawing inferences about the extent of legally proscribed behavior from plaintiff success rates.
Enter the Stephens-Davidowitz study. With his Google-based measure of discrimination in hand, it is a trifling task to rank states or media markets by the resident population’s animosity toward African Americans. Moreover, the powerful linkage between discrimination so measured and voting behavior means that a Voting Rights Act which targeted the worst-offending regions could be defended as a permissible response to the problem of election outcomes that are unconstitutional because of the racial basis for the electorate’s verdict. (Elsewhere I have explained at length why election outcomes are unconstitutional if the winning candidate would not have prevailed but for racially biased votes. For a variety of reasons such constitutional violations cannot be remedied by the courts in litigation about a particular election, but they can, and should, be tackled by Congress through prophylactic legislation such as the VRA.)
A quick glance at Stephens-Davidowitz’s ranking of the states reveals that there is, in fact, a positive correlation between covered status and anti-black prejudice. But the correlation is not impressive. West Virginia ranks first by racial animus but is not covered. Indeed, six of the top ten states by racially charged Google searches fall outside the reach of Section 5. In addition to West Virginia, the uncovered states of Pennsylvania, Kentucky, Michigan, Ohio, and New Jersey have the dishonor of placing in the top 10.
But the more fundamental point is this: Congress can only be expected to craft a coverage formula tailored to the incidence of intentional discrimination if there exists a normatively acceptable and reliable method of ranking geographic regions by their residents’ propensity to discriminate. This undertaking, which not long ago seemed nearly impossible, is now within easy reach of the concerned legislator. (To be sure, the relevant congressional subcommittee would have to commission additional work by Stephens-Davidowitz or others to obtain Google-metrics of discrimination against non-black minorities.)
This past May, the US Court of Appeals for the District of Columbia Circuit rejected the latest constitutional challenge to Section 5 on a 2-1 vote. Judge Stephen Williams, dissenting, would have enjoined the states from enforcing Section 5 unless or until Congress revises the coverage formula. The DC Circuit gave no indication that it was familiar with Stephens-Davidowitz’s work (which was featured in the New York Times shortly after the court’s decision). But I have little doubt that the Supreme Court will take the case and adopt Judge Williams’s position, and I fully expect that at least some of the Justices in the majority will rest their argument on Google. It’s conceivable that portions of Section 5 will be left operable, but I am no longer optimistic.
Should the Supreme Court demand that Congress update the coverage formula, this will be tantamount to invalidating Section 5 outright. In the current political environment, it’s hard to imagine the necessary bipartisan coalition agreeing to a fix.
The Revitalization of Section 2
The implications of Stephens-Davidowitz’s work for Section 2 of the VRA are much brighter. To repeat, Section 2 applies nationally, and requires minority plaintiffs to prove that the challenged electoral structures prevent them from “participat[ing] in the political process and elect[ing] candidates of their choice” on equal terms with other voters. What it means for an electoral structure to have this effect is not entirely clear, and in recent years Section 2 has suffered a string of narrowing interpretations at the hands of an increasingly conservative Supreme Court. The constitutional avoidance canon has been a mainstay of these decisions.
Most Section 2 cases have been brought by plaintiffs seeking the creation of electoral districts in which minority voters would have more influence or representation than they do under the status quo. As a threshold matter, such plaintiffs must show that voting is “racially polarized,” meaning that members of the plaintiffs’ racial group tend to prefer different candidates than do members of other racial groups. However, Section 2 does not require plaintiffs to demonstrate a recent history of unconstitutional racial discrimination by the government so elected, or that the remedy they seek would undo unconstitutionally discriminatory state action or prevent new instances of governmental discrimination going forward. Courts and commentators have therefore doubted whether Section 2 is a “congruent and proportional” response to constitutional violations (the legal standard for enforcement legislation under the Fourteenth Amendment and probably the Fifteenth Amendment).
Stephens-Davidowitz’s work is incredibly important to the future of Section 2, for a couple of reasons. First, it establishes that anti-black animus remains a powerful force in the voting booth today. Across the nation as a whole, “between 6.7 and 10.7 percent of white Democrats did not support Obama because he was black.” This is a shocking finding. In a high-profile presidential election, in which voters have vastly more information about the candidates’ nonracial attributes and policy positions than they do in congressional, state, and local elections, and in which voters are cued to rely on their partisan identities by the ballot itself, roughly 10% of white Democrats defected from their party’s candidate. The effect of racial animus on support for black candidates in down-ballot races is almost surely greater.
Insofar as Section 2’s constitutional justification lies in racial discrimination by the electorate, as I have argued, the findings of Stephens-Davidowitz show that Section 2 responds to a real and substantial constitutional problem.
Stephens-Davidowitz’s research will also help to solve one of the central doctrinal and practical difficulties in litigating Section 2 cases: establishing a nexus between “subjective discrimination” — i.e., decisions that would have been different had the race of persons considered by the decisionmaker been different — and the barrier to minority political participation that’s at issue in the case.
The circuit courts have split on whether Section 2 plaintiffs must trace the electoral inequality at issue to subjective racial discrimination, either by conventional state actors or by the majority-group electorate. Most courts nominally adhere to the causation requirement, but as Professor Jim Greiner observes [PDF], they often subvert it in practice through aberrational burden-shifting rules. In a recent article I argued that Section 2 plaintiffs should be required to make the causation showing, albeit subject to a relaxed evidentiary standard. But, as several readers pointed out, I didn’t say much about what kinds of evidence ought to suffice or how that evidence might be obtained.
The Holy Grail for implementing the causation requirement is a measure of societal discrimination that’s reliable, tied to voter behavior, geographically specific, and low-cost for plaintiffs to produce. With such a measure in hand, the courts could create truly sensible evidentiary presumptions and burden-shifting rules. For example, they might presume that racially polarized voting is caused by racial prejudice in locales that rate “worse than average” on the measure of societal discrimination, but by socioeconomic differences in other areas. (Such presumptions should be rebuttable.)
Stephens-Davidowitz’s Google-based metric of anti-black discrimination fits the bill precisely. It is replicable, it explains otherwise puzzling disparities in voter behavior, it is specific to each of 210 geographic regions, and it appears to be cheap to produce and update. To be sure, there is a lot of additional work that could be done to validate the measure vis-à-vis non-Presidential elections, and to create and validate analogous measures of anti-Latino, anti-Native American and anti-Asian sentiment. Nor is Stephens-Davidowitz’s approach the only promising way to measure or proxy voter discrimination. But his work represents a huge leap forward, and has genuine potential to revolutionize Section 2 litigation.