Racially Polarized Voting

I.  The Purpose of the Racial-Polarization Test

To ground our argument, this Part briefly reviews the evolution of the Gingles framework, focusing on three dominant refrains in jurists’ and commentators’ thinking about the threshold inquiry into polarized voting: manageability, normative diagnosis, and the need to limit reliance on strong racial assumptions.

A.    Origins

The Supreme Court recognized racial vote dilution claims under the Equal Protection Clause in 1971.42 A free-form jurisprudence soon developed in the lower courts, with judges making “intensely local appraisal[s]” of the “totality of the circumstances” bearing on minority political opportunity and then simply pronouncing that the challenged electoral systems did or did not dilute minority voting power.43 Nearly all of the cases concerned at-large elections.44 Appraising this first decade of racial vote dilution law, James Blacksher and Larry Menefee wrote:

Most of the cases concluding . . . that at-large [elections] were constitutional cannot be distinguished analytically from those reaching a contrary result on any basis other than the varying personal political views of the trial and appellate judges who decided them. Some capriciousness is an inherent risk of a standard calling for an “intensely local appraisal” of the “totality of circumstances” of each case.45

This critique led Blacksher and Menefee to propose the racial-polarization test. In lieu of open-ended balancing, courts would simply determine (1) whether the minority community was populous and geographically concentrated enough to compose a majority of an ordinary single-member district, (2) whether the minority community was politically cohesive, and (3) whether the majority community voted sufficiently as a bloc to usually defeat minority-preferred candidates under the status quo electoral system.46 “In terms of certainty and consistency,” Blacksher and Menefee asserted, this inquiry “promises to be nearly as manageable as the population equality rule”47 of Reynolds v Sims,48 in which the Court held that electoral districts must be “as nearly of equal population as is practicable.”49 Key to their manageability claim was that “[t]he problem [of vote dilution would be] observed solely on the basis of voting patterns; there is no need to inquire into the political stories behind the election returns.”50

Blacksher and Menefee’s timing was propitious. In 1980, the Supreme Court rejected the constitutional vote dilution jurisprudence of the 1970s.51 Congress responded by amending § 2 of the VRA, creating a new “results test.”52 The amended § 2 prohibits electoral structures that “result[ ]” in members of a class of citizens defined by race or color “hav[ing] less opportunity than other members of the electorate to participate in the political process and to elect representatives of their choice.”53 Nowhere does the statute or its legislative history provide a clear statement of what it means for a racial minority to have unequal political opportunity.54 All the enacting Congress bequeathed to the courts was a statutory directive to consider the “totality of circumstances”55 and a committee report reciting a nonexhaustive list of factors to be weighed (gleaned from the constitutional vote dilution jurisprudence of the 1970s).56

When amended § 2 first reached the Supreme Court in the 1986 case Gingles, Justice William Brennan, writing for a five-justice majority, sidelined this “totality of the circumstances” analysis in favor of Blacksher and Menefee’s more streamlined inquiry into polarized voting.57 It is now well settled that vote dilution plaintiffs must establish the so-called Gingles conditions at the outset of their case.58 Specifically, plaintiffs must show:

First, [that] the minority group . . . is sufficiently large and geographically compact to constitute a majority in a single-member district. . . . Second, [that] the minority group . . . is politically cohesive. . . . [And third,] that the white majority votes sufficiently as a bloc to enable it—in the absence of special circumstances, such as the minority candidate running unopposed—usually to defeat the minority’s preferred candidate.59

B.    Evolving Conceptions of the Gingles Test

The Gingles Court left some doubt whether satisfaction of the Gingles conditions actually establishes that plaintiffs lack equal political opportunity within the meaning of § 2, or simply that a lack of opportunity (if proved) could be remedied. Put differently, is the Gingles test a liability standard or a test for potential remedies?60

Brennan clearly saw it as a liability standard,61 and many lower courts initially adopted this interpretation.62 This practice cried out for justification because, as Justice Sandra Day O’Connor explained in her concurrence, neither the text nor the legislative history of § 2 offered much support for the plurality’s liability standard.63

Commentators and judges defended the new approach on pragmatic and normative grounds. Echoing Blacksher and Menefee, they argued that the formulaic Gingles standard was needed to make politically sensitive vote dilution cases manageable.64 Professors Lani Guinier and Samuel Issacharoff, for example, emphasized that the easy-to-apply Gingles standard obviated any need for judges to assess governmental responsiveness to the “particularized needs” of the minority community, a quintessentially political issue that was prominent in the constitutional vote dilution jurisprudence of the 1970s.65

Normative defenses of the Gingles standard were tied to the concept of polarized voting, which was believed to signify something important about political equality.66 Depending on the commentator or the moment, “polarized voting” was treated as a reliable indicator of minority exclusion from the normal push and pull of pluralist politics, or of intentional race discrimination against minority candidates by white voters and elites.67

The Supreme Court’s 1994 decision in Johnson v De Grandy68 invigorated the alternative, “potential remedy” reading of Gingles.69 Justice David Souter, writing for the Court, reversed and criticized a lower court that had “misjudged the relative importance of the Gingles factors.”70 Many judges understood De Grandy as a signal to revitalize the totality-of-the-circumstances rubric of 1970s vote dilution law,71 and as Professor Adam Cox and Dean Thomas Miles have shown, the relationship between the satisfaction of the Gingles conditions and a finding of liability weakened in the De Grandy era.72

Even so, some lower courts continued to say that “cases will be rare” in which the Gingles conditions were met but liability not found.73 These courts evidently believed that the Gingles conditions were at least somewhat diagnostic of racial vote dilution.

Souter’s potential-remedy theory of Gingles took a blow in Strickland, the Supreme Court’s most recent statement about the Gingles framework.74 The question in Strickland was whether the first prong of Gingles is to be read formally (such that plaintiffs must compose a literal, numeric majority of voting-age citizens in the proposed remedial district) or functionally (to allow claims by minority voters who, though composing a numeric minority of the remedial district, would be able to control it in coalition with sympathetic white voters).75 Justice Anthony Kennedy, writing for the Strickland plurality, adopted the formal approach.76

In dissent, Souter bemoaned that states could now comply with § 2 only by segregating minority voters into majority-minority districts, rather than by drawing districts to empower cross racial coalitions.77 Kennedy replied that Souter fundamentally misunderstood the plurality opinion.78 Certainly a state could comply with § 2 by creating crossover districts in lieu of majority-minority districts.79 After Strickland, crossover districts voluntarily adopted by a state may forestall § 2 liability ex ante or remedy liability ex post, yet they cannot satisfy the first prong of Gingles. Thus does Strickland sever Gingles from potential remedies.

What then is the point of the Gingles-Strickland majority-minority requirement? Per Kennedy, it is not to hem in the states’ compliance efforts, but simply: (1) to limit the reach of § 2, lest it “unnecessarily infuse race into virtually every redistricting, raising serious constitutional questions”;80 and (2) to reduce the number of cases in which courts end up in the “untenable position of predicting many political variables and tying them to race-based assumptions,”81 or otherwise making “highly political judgments.”82

The notion that the Gingles conditions should be used to limit the geographic reach of § 2 is also reflected in the Strickland plurality’s discussion of white bloc voting. Kennedy chided the state for conceding white bloc voting, expressing “skeptic[ism]” that this condition could be satisfied as to districts where 20 percent of white voters support minority-preferred candidates.83 Though Kennedy wrote for only three justices, it appears from oral argument that at least five justices were open to the idea of numeric cutoffs for group cohesion.84 Suggested cutoffs ranged from 15 percent to 40 percent crossover voting, or, equivalently, 60 percent to 85 percent of whites voting against the minority-preferred candidate.85 The intuition here is that courts ought not to get involved in vote dilution cases unless preference polarization as revealed by votes is extreme. Where the preference divergence is only moderate, dilution problems will generally be less severe, both because representatives elected without decisive minority support are likely to be less horrible (from the point of view of minority voters) than in jurisdictions where preference polarization is severe and because there is some chance that successful biracial or multiracial coalitions will form.86

If the Gingles test is to be used to curtail vote dilution litigation, as Strickland suggests, then the test must be normatively diagnostic. It would be arbitrary and unlawful for the Court to foreclose consideration of the liability standard prescribed by statute without reference to whether that standard is likely to be satisfied.87 Kennedy would not disagree. The Gingles conditions, he wrote in Strickland, are tools “to help courts determine which claims could meet the totality-of-the-circumstances standard for a § 2 violation.”88

To summarize, in the post-Strickland world, the Gingles test must be objective and clear-cut, and the test must also be reasonably well calibrated to filter strong from weak claims—and reasonably free of strong racial assumptions. It should be noted that although this conception of the Gingles test was crystallized in Strickland, the crystallization was hardly revolutionary. Many judicial opinions in the years between Gingles and Strickland are premised on the idea that the Gingles test is or should be clear-cut, easy to administer, and normatively diagnostic.89 Similarly, Strickland’s notion that the VRA should be interpreted so as to minimize reliance on race-based assumptions has an ample lineage, dating to the early 1990s.90 Indeed, just a few years before Strickland, the Supreme Court reversed a lower court that had, in Kennedy’s view, improperly assumed political commonality among voters who shared the same ethnicity but were geographically and socioeconomically remote from one another.91

II.  Judicial Practice in the Lower Courts

As the evolution from Gingles to De Grandy and then Strickland suggests, the Supreme Court justices have not been of one mind about the extent to which the Gingles factors are diagnostic of liability. One would expect to find similar equivocation in the lower courts.

By contrast, the notion that the Gingles conditions serve an essential manageability function has long been standard wisdom among courts and academics alike.92 Equally uncontroversial is the proposition that courts should avoid making racially essentialist assumptions.93 The Equal Protection Clause generally prohibits state actors from relying on assumptions about the preferences, abilities, or behaviors of members of a class defined by race or ethnicity, even if the assumptions have some grounding in reality.94

But as this Part explains, the lower courts have not developed a crisp, objective test for racially polarized voting, let alone one that can be applied without recourse to strong racial assumptions. The pattern of judicial votes in vote dilution cases does not bear out the hypothesis that judges are substantially constrained by the Gingles framework, and a close look at the doctrine suggests that judges have struggled to reconcile the desire for an objective test with the desire to avoid racial assumptions.

A.    Judicial Votes

Reflecting and reinforcing conventional wisdom about the rule-like, constraining nature of the Gingles test, an influential paper by Professor Cox and Dean Miles presents judicial voting at the Gingles and totality-of-the-circumstances stages of § 2 cases as a test of the hypothesis that doctrinal “rules” constrain judges more than “standards.”95 Cox and Miles show that Democratic and Republican judges vote to find the Gingles conditions satisfied at roughly the same rates.96 However, in the pre-1994 period, Democratic judges were about 45 percentage points more likely than Republican judges to vote for liability conditional on passing the Gingles test.97 In the post-1994 period, the Democratic/Republican gap in liability votes “disappeared.”98 Cox and Miles argue that a changing mix of cases gave Democratic judges partisan incentives to vote against liability in the post-1994 period.99 These judges adjusted their behavior at the unconstrained second stage of vote dilution cases (the totality-of-the-circumstances stage) but not at the rule-like Gingles stage.100

Cox and Miles’s Gingles-constraint hypothesis is undermined, however, by two very important results that appear in their tables but are barely mentioned in the text of their article. White and African American judges diverge more from one another when voting on the Gingles test than when voting for or against liability conditional on the Gingles test being met.101 Similarly, the effect of an African American copanelist on voting by other judges is larger at the Gingles stage than at the totality-of-the-circumstances stage.102 (It is not as if black judges are uniquely unconstrained by Gingles.) These results suggest that the Gingles “rules” are, if anything, less constraining than the totality-of-the-circumstances “standard” for liability.103

Cox and Miles also downplay a very plausible legal explanation for the huge pre-1994 difference in Democratic and Republican votes for liability, conditional on the Gingles factors being met.104 During this period, it was open to lower courts to adopt either the liability-standard interpretation of Gingles or the potential-remedy interpretation.105 The pattern uncovered by Cox and Miles would occur if Democratic judges initially adopted the liability-standard approach (in line with Justice Brennan’s plurality opinion) and if Republicans went for the potential-remedy gloss (in keeping with Justice O’Connor’s concurrence). In 1994, the Supreme Court in De Grandy reversed a district court for treating Gingles as a liability standard.106 Cox and Miles’s demonstration of a marked post-1994 change in liability-stage voting by Democratic judges may reflect nothing more than De Grandy’s rejection of the liability-standard reading of Gingles.

B.    Doctrine and Practice

Cox and Miles’s results regarding the direct and panel effects of judge race are surprising if one believes the Gingles conditions to be rule-like. But as this Section explains, the applicable law leaves judges with broad discretion under the second and third Gingles prongs.

The nuts and bolts of the racial-polarization test represent something of a black box. The Supreme Court has said little about them since Gingles, and the leading casebooks treat the matter briefly or not at all.107 To get a handle on judicial practice, one of us (Elmendorf) read dozens of prominent court of appeals opinions. Then, to obtain a clearer sense of the law as it is experienced and applied by fact finders, we asked law students to code a random sample of opinions issued by district courts following bench trials in racial vote dilution cases. Stratified by time period, our sample includes twenty-five cases from the Gingles to De Grandy period, twenty-seven from the De Grandy to Strickland period, and the universe of all seventeen available district court opinions following bench trials in the post-Strickland period. Appendix B explains the details of our sampling and coding protocol.

Our aim in this Section is to present an evenhanded and succinct summary of doctrine and practice under Gingles. (In the interest of readability, we relegate most quantitative summaries of student codings to footnotes.) Three themes emerge from our review. First, most courts have been reluctant to establish bright-line rules about central questions such as the levels of voting cohesion needed to establish “legally significant” preference polarization, the identification of “minority candidates of choice,” and what constitutes “usual defeat” of such candidates. Second, judges disagree about many fundamental matters, including the legally relevant dimension of political preferences; whether the voting-power (“usual defeat”) component of the Gingles test should be understood in terms of observed outcomes, or instead in terms of opportunities that exist in principle; the weight to be given to polarization evidence from primary as opposed to general elections; the proper geographic scale of the analysis; and which candidates qualify as “minority candidates of choice.” Third, the courts seem torn between relatively simple, mechanical polarization inquiries, which often depend on race-based assumptions, and more-nuanced approaches, which entail considerable attention to the very thing Blacksher and Menefee expected judges to ignore: “the political stories behind the election returns.”108

This Section might be seen as ammunition for the Strickland Court’s campaign to establish restrictive vote-share cutoffs for minority cohesion and white bloc voting, or as a demonstration of the need for categorical rules about which elections to include in or exclude from the polarization analysis. But those are not the right lessons. The lower courts have not been indifferent to manageability norms or to norms against race-based assumptions. Rather, as we explain in Part III, their failure to settle on clear, prescriptive rules at the Gingles stage (free of race-based assumptions) simply reflects the impossibility of this assignment, given ongoing normative disagreements about the meaning of racial vote dilution and the convention of basing racial-polarization findings on votes cast in actual elections.

1.   Preliminaries: the privileged status of election data.

If there is any well-settled convention under the second and third prongs of Gingles, it is this: “[T]he issue of political cohesiveness [shall] be judged primarily on the basis of the voting preferences expressed in actual elections.”109 Defendants have occasionally tried to rebut vote-share evidence with survey or other data, but these efforts have universally failed.110 Anecdotal testimony is sometimes offered to reinforce or illuminate the voting data, but estimates of candidates’ vote shares by racial group in usual elections have been the center of attention since Gingles.111

The “usual” caveat is important: the Gingles Court recognized that voting patterns in certain elections may be anomalous because of special circumstances, such as the presence of an exceptionally strong incumbent or the lack of a challenger.112 But the Court (like Blacksher and Menefee) apparently assumed that once these anomalous elections are excluded, everything a judge needs to learn about minority cohesion and white bloc voting can be learned from typical white and minority vote shares in other elections.113

2.   Preference polarization.

Courts typically assess the legal significance of preference polarization in two steps, asking first whether the minority community is politically cohesive (the second prong of Gingles) and then whether the white majority votes as a bloc against the minority (the third prong).114

a) Minority preferences.  In a widely noted book published in 1992, Bernard Grofman, Lisa Handley, and Richard Niemi asserted that Gingles had settled the question of how to prove minority cohesion.115 Evidence of a statistically significant correlation between voter race and vote choice was both necessary and sufficient to establish minority cohesion.116 But it is of course true that statistical significance does not imply substantive importance, and notwithstanding Grofman, Handley, and Niemi’s interpretation of Gingles, courts frequently ask whether minorities vote by large or “landslide” margins for minority-preferred candidates.117

The landslide approach was urged by expert witness Professor Allan Lichtman and prominent litigator J. Gerald Hebert in a 1993 article.118 They proposed that if 60 percent or more of minority voters typically vote for the minority candidate in biracial, two-candidate elections, minority cohesion should be presumed as a matter of law.119 If minority in-group voting falls below 60 percent, then courts should consider whether “local conditions” (such as a lack of “viable” minority candidates) might explain the apparent lack of cohesion.120

Judicial practice comports with Lichtman and Hebert’s suggestion in one important respect: findings about minority cohesion—or its absence—are almost universally based on rates of coethnic voting by plaintiff-race citizens in elections contested by plaintiff-race candidates. There were thirty-nine cases in our sample in which minority cohesion was disputed at trial and the court reached the issue. In all but one of these thirty-nine cases, the court grounded its determination exclusively or primarily on data from elections with a minority-race candidate.121

Though judges ruling on minority cohesion often emphasize the percentage of minority voters who supported minority-race candidates,122 the lower courts have been very reluctant to establish quantitative cutoffs for what is, or is not, legally significant minority cohesion. (This is despite Justice Thomas’s early warning in Holder v Hall[fn value="123"]512 US 874 (1994). that the establishment of cutoffs would mitigate some of the problems with racial vote dilution law.)124 A few courts have deemed in-group voting rates of 60 percent to 70 percent presumptively cohesive.125 Yet of the thirty-nine pertinent cases in our sample, there appears to be only a single instance in which a court even gestured toward a minimum level of voting cohesion, or toward a minimum gap between minority and white vote shares, that would be required to satisfy the second prong of Gingles.126

Considered in view of manageability concerns and the Supreme Court’s injunction against race-based assumptions, the lower courts’ handling of the second (minority cohesion) prong of Gingles invites two objections. First and most obviously, the courts have not set numeric cohesion cutoffs, as Strickland suggests that they should.127 Second, the de facto definition of “minority political cohesion” as “consistent voting by large margins for coethnic candidates” rests heavily on racial assumptions. It presumes that minority voters who share the same political views will manifest their commonality by obdurately supporting candidates of their own race, essentially ignoring the issue positions and the strengths and weaknesses of each minority candidate.128 It also presupposes that, so far as minority voters are concerned, the candidates who face off against plaintiff-race candidates are just an undifferentiated mass, one no better than the next.129 Whereas the law generally favors race blindness, the conventions that guide judicial fact-finding about minority political cohesion effectively assume that voters—if cohesive—see only race when choosing among candidates.130

b) White preferences.  The plurality opinion in Gingles stated that white bloc voting is “legally significant” if “whites vote sufficiently as a bloc usually to defeat the minority’s preferred candidates.”131 Read literally, this implies that white preferences do not have to be estimated. Rather, after determining that a minority community is cohesive, the court simply has to establish which candidates count as “minority preferred” (that is, “minority candidates of choice”) and see whether they won or lost.

In point of fact, however, expert witnesses report and courts make findings on the share of the white vote that went to minority-preferred candidates.132 This coheres with the sentiment that the justices expressed in Strickland—to wit, that substantial levels of white crossover voting may defeat a vote dilution claim, even if the white voters do not cross over in sufficient numbers to usually elect the minority’s candidate of choice.133 (Another way to put this: preference polarization must be fairly severe for a claim to pass through the Gingles sieve.)

However, in none of the fifty-nine cases in our sample in which the district court passed on a dispute about white bloc voting did the court state or follow a bright-line rule about levels of white crossover voting that would defeat a vote dilution claim as a matter of law.134 Further reading turned up a few such cases beyond our sample,135 but the vast majority of courts have refused invitations to set, as a matter of law, quantitative thresholds that preclude (or require) a finding of legally significant white bloc voting.136

Some courts have sought to distinguish legally significant from legally inconsequential white preferences by asking not simply whether white and minority voters want different things in the political arena but whether race causes the preference divergence.137 These courts have been imprecise in explaining the causation requirement, sometimes talking about the effect of candidate race on vote choice and in other cases suggesting that the requirement might be about the “effect” of voter race on the voter’s choice of candidates.138 In the First and Fifth Circuits, causation must be shown;139 elsewhere, it is only one consideration among many to be weighed, either as part of the Gingles analysis140 or at the totality-of-the-circumstances stage.141 Even where it is a formal requirement, however, the causation test appears to do little to constrain judicial discretion.142

c) What is a usual (probative) election?  As we noted above, the Supreme Court in Gingles distinguished “usual” elections from those with “special circumstances,”143 positing that in some atypical cases, racial-group vote shares may not reflect underlying political differences or provide an accurate picture of minority voting power. Given manageability concerns, one might expect the courts to have categorically excluded evidence from elections with “special circumstances” and to have defined the class of special-circumstances elections with reference to discrete, objective characteristics.144

But instead of firm rules, most courts have opted for loose guidelines tied to what the courts sometimes call “probativeness.”145 For example, interracial elections are generally regarded as more probative than monoracial elections,146 and elections to the governmental body at issue in the case (“endogenous” elections) are customarily given more weight than elections to other governmental bodies (“exogenous” elections).147

Within this rubric, the courts continue to disagree about some very basic matters. For example, should primary elections be included in the analysis?148 If so, should they receive more or less weight than general elections? Most courts include primary elections149 and sometimes give them particular emphasis, on the theory that voting patterns in such elections are uncontaminated by partisan cues and therefore are more truly reflective of racial polarization in the community.150 On the other hand, primary elections are usually low-turnout affairs, and judges have been wary of inferring polarization from low-turnout elections, reasoning that in such contests the electorate may be unrepresentative of the entire political community.151 Other judges downplay primary election voting patterns because of the Supreme Court’s statement that “minority voters are not immune from the obligation to pull, haul, and trade to find common political ground.”152 The thinking here is that a candidate who splits the minority vote in a primary still ought to qualify as a candidate of choice if she can muster support from a broad coalition of white and minority voters in the general election.153 In short, not only are there no clear rules for excluding or weighting primary election data, but there is not even a consensus about whether polarization findings from primary elections should be up-weighted or down-weighted relative to findings from general elections.154

The fluidity of the special-circumstances doctrine and the probativeness rubric also leaves fact finders with broad leeway to heed (or ignore) “the political stories behind the election returns.”155 Judges have discounted voting patterns in particular elections because of such “story-based” determinations as:

• The minority candidate was “an attractive, well-known incumbent whose appeal cut across racial lines.”156
• The minority candidate was “weak.”157
• The appearance of cohesive voting could have been due to a “friends and neighbors” effect—that is, voters supported a candidate because he was the hometown guy, rather than because of meaningful political commonality within a racial group.158
• Weak minority turnout suggested disinterest in the election.159
• The election concerned a ballot measure that was not shown to be of particular interest to the minority community.160
• The minority candidate was backed by a local, predominantly white political faction.161
• The minority candidate was “an ex-pro athlete.”162
• The election postdated the filing of the lawsuit, and white elites may have organized to elect a minimally acceptable minority candidate in the hopes of defeating the lawsuit.163

Notably, no court outside of the Fourth Circuit requires plaintiffs to estimate racial voting patterns in all or a representative sample of elections in the relevant time period.164 What counts as an “unusual” election is therefore in the eye of the beholder and is not determined with reference to the normal range of variation in local elections.

3.   Voting power (lack of opportunity to elect).

Recall that under Gingles, plaintiffs must establish both that the distribution of political preferences is legally significant (“preference polarization,” in our terminology) and that the minority community lacks the opportunity to elect its preferred candidates (“voting power”).165

The vast majority of courts have addressed the voting-power question by, in essence, trying to predict the future win rates of minority candidates of choice.166 Generally this prediction is implicit, grounded on past experience. The court first determines which of the elections for which data were presented to the court were contested by the minority community’s candidates of choice; it then gauges the probativeness of each of these elections, taking account of any special circumstances; and it concludes with a gestalt judgment about whether the actual or anticipated win rate is low enough to satisfy the third prong of Gingles.167

A few courts have resisted this approach, however, arguing that the success rate of minority candidates of choice is less important than whether those candidates could be elected if voting-eligible minority citizens registered, turned out at high rates, and voted very cohesively for the minority-preferred candidate.168 These courts sometimes reference the Supreme Court’s decision in De Grandy, which emphasizes that § 2 guarantees equality of opportunity for minority voters, not success for minority-preferred candidates.169 Some courts have implemented this opportunity-not-outcomes approach in a fact-intensive manner, trying to make realistic judgments about feasible rates of minority turnout and voting cohesion.170 Other judges deem minority voters to have the opportunity to elect as a matter of law in any district where they compose a majority of the voting-eligible population.171

Because the great bulk of lower courts have approached the voting-power question by assessing the win rate of minority candidates of choice, the balance of this Section focuses on how courts identify such candidates and define “usual” defeat.

a) Identifying the minority’s “candidates of choice.”  Judges almost universally treat any plaintiff-race candidate who won a plurality of the minority vote as a candidate of choice.172 But judges have been much more reluctant to use vote shares alone to classify white candidates as minority candidates of choice.

In the Third and Tenth Circuits, defendants may present evidence from white-versus-white elections to rebut the plaintiffs’ showing of minority cohesion and white bloc voting in interracial elections.173 However, before crediting these rebuttal data, the judge must undertake a “detailed, practical evaluation” of whether particular white candidates are, “as a realistic matter,” champions of the minority community.174 Notice that this imports a candidate-level version of the 1970s-era government-responsiveness inquiry into the racial-polarization analysis. (Recall Professor Guinier’s and Professor Issacharoff’s arguments that the whole point of the Gingles framework was to obviate the need for any such inquiry.)175

The Second and Ninth Circuits have forcefully rejected the Third and Tenth Circuits’ “subjective” definition of “minority candidate of choice,” deriding the “detailed, practical evaluation” as “a dubious judicial task, and one that can degenerate into racial stereotyping of a high order.”176 The Second and Ninth Circuits’ “objective” alternative is to credit white-versus-white elections, but only if at least one candidate was strongly preferred to the others by minority voters.177 The Fourth Circuit has gone one step further and now requires plaintiffs to present data from all or a “representative sampling” of elections in the defendant jurisdiction during the relevant time period.178 Any candidate who would have won if only minority-cast ballots were counted is presumed to be a minority candidate of choice.179

Even more striking than the diversity of legal doctrine across the circuits with respect to white-versus-white elections is its instability within circuits—and even within judicial opinions. Thus the Ninth Circuit, in the very same case in which it vehemently rejected the Third and Tenth Circuits’ subjective definition of “minority-preferred candidate,” held that voting data from elections without a minority-race candidate generally deserve less weight in the polarization analysis.180 How much less weight? The court did not say, and we do not see how one could make this judgment without assessing whether the white-versus-white elections were fought over issues that divided the white and minority communities—a question that can be answered only through the kind of contextual, stories-behind-the-elections inquiry favored in the Third and Tenth Circuits. The Ninth Circuit is not an outlier. Judges in the Second and Sixth Circuits, which nominally follow the objective approach, also give an indeterminate amount of additional weight to polarization evidence from interracial elections.181

Even the Fourth Circuit has wavered. Initially, it espoused a version of the subjective approach.182 When the Fourth Circuit subsequently created the representative sampling requirement, the court spelled out bright-line vote-share rules about which candidates are “minority preferred” in elections in which several candidates are chosen concurrently.183 But the court also allowed that “individualized assessment[s]” of particular candidates may be warranted in certain borderline cases.184 A decade later, the Fourth Circuit held that with respect to elections in which no candidate received a majority of the minority vote, the fact finder should consider “testimony from political observers and the candidates themselves” to determine whether any of the candidates “can be fairly considered a representative of the minority community.”185 This is the crux of the subjective approach, as practiced in the Third and Tenth Circuits.

But it is not as if the courts are generally converging on the subjective approach. We see just as much movement in the other direction. The Eighth Circuit, which initially embraced the subjective approach,186 subsequently upheld a district court’s decision that understood a “minority candidate of choice” as “any candidate who gets enough votes to be elected if only minority ballots are counted.”187 A similar transition seems to have occurred in the Eleventh Circuit.188 In short, the courts are going in circles.

b) Quantifying “usual defeat.”  The question of what counts as “usual defeat” (or its obverse, “opportunity to elect”) within the meaning of Gingles has received little elaboration. Courts variously ask whether the minority community has a “fighting chance,”189 a “good chance,”190 or a “reasonable likelihood”191 of electing its candidates of choice, or whether minority voters constitute an “effective political force.”192 It is very hard to pin down what judges actually mean by these formulations. One of us tried to code our random sample of district court opinions for three plausible conceptions of what is required to establish “usual defeat”: (1) a win rate of less than 50 percent; (2) a win rate of much less than 50 percent, such that victories are rare; or (3) a win rate resulting in the election of a disproportionately small number of minority candidates of choice relative to the minority’s population share. In 90 percent of the cases, the coder was unable to say which if any of these conceptions the court was using.193 “Usual defeat” is a legal black box.

4.   A note on geographic scale.

Some courts assess racial polarization at the geographic scale of the polity as a whole, or at least over the same region that the court uses to gauge proportionality between the number of minority opportunity districts and the minority’s population share.194 (Proportionality is often a central consideration at the totality-of-the-circumstances stage of vote dilution cases.)195 Other courts focus more narrowly on the plaintiff’s electoral district.196 In cases about at-large elections, there is no daylight between these approaches; the plaintiff’s electoral district is geographically coterminous with the polity.197 But in the more typical case today, addressing the configuration of single-member districts rather than the choice between districted and at-large elections, the scale issue may be very consequential. It determines whether minority voters in locally polarized subsets of a generally nonpolarized state or city can bring a vote dilution claim. It may also affect whether “opportunity districts”198 for minorities in a relatively nonpolarized area can be used to remedy dilution in more-polarized areas.199

C.    Summary

The racial-polarization test falls far short of its manageability aspirations. As things stand today, there are no established quantitative cutoffs to distinguish polarized from nonpolarized communities, no clear-edged rules about which elections to include in the polarization analysis and how to weight them, and no settled understanding of what constitutes “usual defeat” or “lack of opportunity to elect.” Moreover, racial assumptions play critical roles in judicial fact-finding about minority cohesion and in the identification of minority candidates of choice. (Hidden racial assumptions are also baked into the statistical tools for estimating candidates’ vote shares by racial group. More on this in a moment.)200

To be sure, some broad guidelines are well established: interracial elections are more probative than same-race elections, and endogenous elections are more probative than exogenous elections. But these guidelines hem in judicial discretion only at the margins.201 Given the general directive to weigh voting data in view of their “probativeness,” taking account of pertinent “special circumstances,” fact finders are encouraged—and in some circuits, required—to pay close attention to “the political stories behind the election returns.”202

Yet it is not as if the manageability aspirations have been lost on the courts. We see them manifested in, for example, the Second, Fourth, and Ninth Circuits’ “objective” definitions of minority candidate of choice;203 in the Fourth Circuit’s “representative sampling” requirement;204 in the occasional attempts to define minority political cohesion or white bloc voting using quantitative thresholds;205 and in the occasional pushback against liberal use of the special-circumstances doctrine.206 The convention of determining minority cohesion through a rote analysis of minority vote shares for coethnic candidates (without paying much attention to the particulars of those candidates and their opponents) can also be understood as one means by which courts have tried to accommodate concerns about manageability.207

Nor has the injunction against race-based assumptions been ignored. Most courts have refused to embed into law the notion that only minority-race candidates can be candidates of choice for minority communities.208 And some prominent judges have rejected doctrines that make candidate-of-choice determinations depend on anecdotal testimony and the judge’s prior beliefs about what minority voters want.209

The puzzle of racial-polarization law is not that the Supreme Court’s manageability and racial-assumption concerns have been ignored but that the various judicial efforts to build doctrine responsive to these concerns have proved so halting. That is to say, there has been no general movement toward bright-line preference-polarization cutoffs; toward the representative sampling requirement; toward an objective, race-neutral definition of the “minority candidate of choice”; toward a conceptually transparent definition of voting power (lack of opportunity to elect); or toward a limiting construction of the special-circumstances doctrine. This is in marked contrast to the lower courts’ development of the first prong of Gingles; nearly all courts anticipated the Supreme Court’s decision in Strickland, holding that plaintiff-race voters must compose a literal, numeric majority of the proposed remedial district.210

Why has the law of racial polarization not developed in keeping with the Supreme Court’s expectations? We turn to this question next.

III.  The Fundamental Problems

As we have seen, the racial-polarization test is supposed to be informative about preferences and power. Polarized voting per Gingles signifies that voters within a racial group want the same things in the political sphere (preference cohesion within groups);211 that voters in different racial groups want different things (preference polarization between groups);212 and that preference polarization coupled with the defendant jurisdiction’s demographics and electoral system means that minority voters have insufficient voting power—that is, insufficient opportunity to elect their candidates of choice.213

The racial-polarization inquiry is also supposed to be free of racial assumptions.214 The judge’s task is to learn whether the racial groups are in fact relevantly cohesive—and whether particular candidates were in fact the minority community’s candidates of choice—without stereotyping minority or white voters in the process. But as this Part explains, there are three massive obstacles to the development of a polarization test that is at once objective and consistently applied, uncontaminated by “prohibited” racial assumptions, and diagnostic of preference polarization and voting power.

The first fundamental problem, addressed in Part III.A, is the lack of an agreed-upon normative definition of racial vote dilution. Several competing theories with very different implications for the Gingles test each have some support in the case law.

The second fundamental problem, addressed in Part III.B, is that vote shares by racial group are an undependable proxy for latent political preferences, owing to strategic behavior on the part of candidates, other political elites, and voters. It is a huge stretch to assume, in line with the Gingles Court and many judges and commentators since, that the average degree of polarization in typical elections for the government body at issue is highly probative of polarization in underlying political preferences. It is equally implausible to assume that a candidate’s minority vote share signals how close she is to being the ideal candidate of the minority community.

The third problem, which is the subject of Part III.C and Appendix A, is that the estimation of candidates’ vote shares by racial group from precinct-level vote totals requires expert witnesses to make racial-homogeneity assumptions similar to those the Supreme Court has deemed “prohibited.”215

Notice that the second and third problems are by-products of the convention of basing polarization determinations on “voting preferences expressed in actual elections.”216 These problems could in principle be solved by using other kinds of evidence to screen vote dilution claims, as we explain in Part IV.

A.    Normative Theory and the Diagnostic Value of the Polarization Test

Professor Guinier once quipped that the VRA is “a statute in search of a theory.”217 Her observation still rings true today. Several competing theories of racial vote dilution find some support in the case law, and to date the Supreme Court has declined to choose among them. This much is well understood in the legal academy, but legal scholars have largely failed to consider how normative disputes color the racial-polarization analysis. (It is telling that the leading casebook notes only one point of connection between the second and third prongs of Gingles and normative disagreements about the meaning of racial vote dilution.)218

The separate opinions of Justices Brennan, O’Connor, and White in Gingles seeded three of the competing normative theories, which we call, respectively, the proportional representation, coalitional-breakdown, and voter-discrimination accounts of vote dilution. Justice Kennedy’s opinion for five justices in LULAC gestures toward yet another theory, still dim in its outlines, which we call the LULAC theory. Part III.A.1 briefly describes these four theories, and Part III.A.2 draws out their conflicting implications for the racial-polarization test.

Most judicial opinions in vote dilution cases are nominally atheoretical,219 but it does not follow that the theories are inconsequential. To the extent that a judge exercises discretion when applying the Gingles test, that discretion is likely informed by whatever theory the judge finds intuitively congenial, even if she does not articulate it.

1.   Four theories of racial vote dilution.

a) Proportional representation theory.220   Brennan’s opinion in Gingles embraced Blacksher and Menefee’s conception of vote dilution,221 which O’Connor fairly characterized as “an entitlement to roughly proportional representation [for geographically and politically cohesive minority groups] within the framework of single-member districts.”222 On this view, voting power is defined in terms of the opportunity to elect ideally preferred representatives, and the fairness benchmark is proportionality: the ratio of the number of districts that provide the minority community with this opportunity to the total number of districts should roughly equal the ratio of the minority population to the total population.223

Subsequent Supreme Court decisions, most notably De Grandy and LULAC, emphasize (without making decisive) whether the minority community has the opportunity to elect candidates of its choice in rough proportion to its numbers.224 As for the lower courts, Professor Ellen Katz and her coauthors reviewed all published § 2 opinions through 2005, and they report that liability rulings in the eighteen lawsuits in which courts “made a finding on proportionality” were very highly correlated with that finding.225

b) Coalitional-breakdown theory.226   The coalitional-breakdown theory begins with the premise that, in a system of plurality-winner elections, there is nothing abnormal about a political minority being unable to elect its ideally preferred candidates.227 Political minorities obtain representation instead by banding together with others in an umbrella coalition that competes for control of the legislative body.228 A political minority’s vote is “diluted” if and only if it lacks the opportunities to participate in coalitional politics that political minorities normally enjoy. Racial vote dilution occurs when a racial minority that is also a political minority lacks such opportunities.

Rooted in the foundational 1970s cases Whitcomb v Chavis229 and White v Regester,230 as well as Justice Thurgood Marshall’s dissent in City of Mobile, Alabama v Bolden231 (the case that Congress overruled in 1982232 ), the coalitional-breakdown theory found expression in Gingles through O’Connor’s concurrence.233 She further advanced the theory in Georgia v Ashcroft,234 a case that arose under § 5 rather than § 2 of the VRA.235

The Court’s post-Georgia decisions in LULAC and Strickland strongly suggest, however, that plaintiffs may not use the coalitional-breakdown theory as a sword in § 2 cases. Per LULAC, vote dilution claims cannot be predicated on mere lack of “influence,” as opposed to the inability to elect candidates of choice.236 Similarly, the Strickland plurality declares, “Nothing in § 2 grants special protection to a minority group’s right to form political coalitions [with majority-race voters].”237

Yet the coalitional-breakdown theory may still be available to defendants as a shield.238 Referencing Georgia’s discussion of “influence” and “crossover” districts, the Strickland plurality states, “Much like § 5, § 2 allows States to choose their own method of complying with the Voting Rights Act.”239 The arguable implication is that defendants can escape liability under § 2 by structuring electoral systems so as to integrate minority groups into competitive political coalitions. The VRA allows state and local governments to make the “political choice of whether substantive or descriptive representation is preferable.”240

The coalitional-breakdown theory nicely rationalizes a number of lower court decisions that reject liability even though the minority group was unable to elect a roughly proportional number of its candidates of choice and more majority-minority districts could have been drawn. Courts finding for defendants notwithstanding a lack of proportionality have often pointed to circumstances such as robust mobilization of minority voters and rough parity in turnout rates, minority participation in candidate slating, nomination and election of minority-race candidates (even if they are not the minority group’s candidates of choice), and government responsiveness to the particularized interests or needs of the minority community.241 Viewed together, such facts tend to indicate that the minority group has been incorporated into the normal push and pull of coalitional politics, even if it cannot usually elect a roughly proportional number of its ideal candidates.

c) Voter-discrimination theory.242  In contrast to the proportional representation and coalitional-breakdown theories of vote dilution, the voter-discrimination theory emphasizes the reasons why minority-preferred candidates lost. The core issue is disparate treatment: whether white voters give less support to minority candidates than to otherwise-similar white candidates.243

Justice Byron White’s concurring opinion in Gingles launched the voter-discrimination theory.244 White joined most of Brennan’s opinion, but he forcefully rejected Brennan’s premise that “there is polarized voting if the majority of white voters vote for different candidates than the majority of the blacks, regardless of the race of the candidates.”245 White surmised that Brennan’s test would require a finding of polarization if whites vote as a bloc for Republicans and blacks for Democrats, even if whites are perfectly willing to support black Republicans.246 “This is interest-group politics,” he lamented, “rather than a rule hedging against racial discrimination.”247

Riffing on White’s opinion, prominent judges on the First, Second, Fifth, and Eleventh Circuits have further developed the voter-discrimination theory.248 But most courts have said that while evidence of voter discrimination deserves some weight249 —maybe a lot of weight250 —subjective discrimination by voters (or conventional state actors) is not a necessary condition for liability under § 2.251 The notion of giving only some, rather than decisive, weight to evidence of voter discrimination seems more in keeping with the coalitional-breakdown theory, as voter discrimination is one of several factors that could make it difficult for racial minorities to participate in normal coalitional politics.252

d) The LULAC Theory?253   Breaking with previous vote dilution jurisprudence, the Supreme Court in LULAC held that Texas’s reconfiguration of an existing majority-minority district violated § 2, even though Texas did not change the number of majority-minority districts, which remained proportional to the minority’s population share, and even though the district court found that the substantially reconfigured district was more likely to elect a minority candidate of choice than the previous district.254 The reconfigured district was “noncompact,” wrote Kennedy, because it combined two geographically and socioeconomically distinct Latino communities.255 Because of this, he continued, it should not count in the proportionality analysis.256 But the really revolutionary idea came next. Even assuming that the shortfall from proportionality was “insubstantial,” that “would not overcome the other evidence of vote dilution for Latinos in District 23.”257

Whereas the proportional representation, coalitional-breakdown, and even, arguably, the voter-discrimination theories focus on aggregate opportunities for the minority group throughout the defendant jurisdiction, the LULAC Court took the view that a particular geographic cluster of minority voters could suffer dilution irrespective of the representational opportunities available elsewhere to the minority group.

Yet since De Grandy, it has been settled that majority-minority districts need not be created for every “potential majority” cluster of minority voters.258 So what exactly was the problem with Texas’s dismantling of District 23? Commentators and lower courts have puzzled over this question. One possibility is that LULAC turned on the appearance of intentional discrimination. Kennedy wrote that the state’s dismantling of a majority-minority district whose voters were about to unseat a disliked incumbent “b[ore] the mark” of intentional discrimination.259 On this reading, § 2 prohibits not only state actions that are in fact intentionally discriminatory but also those for which there is some visible (but perhaps equivocal) evidence of intentional or disparate treatment discrimination.260

LULAC can also be read for the proposition that § 2 protects all “naturally occurring” majority-minority districts—that is, those likely to be drawn by a politically neutral redistricter applying traditional criteria such as compactness, respect for communities of interest, and minimization of political subdivision splits.261 Building on this reading of LULAC, Judge Easterbrook wrote for the Seventh Circuit that the failure to draw a majority-minority district violates § 2 only if the defendant created fewer compact majority-minority districts than a politically neutral computer algorithm likely would have drawn.262

The “appearance of discrimination” and “naturally occurring opportunity district” glosses on LULAC are close kin, because reasonable observers may fairly wonder whether a state has acted for discriminatory reasons if it willfully departs from traditional districting principles and thereby fragments a cohesive minority community into districts in which it is powerless.

2.   Implications for the polarization test.

For the Gingles factors to speak to liability, they must say something about whether the distribution of political preferences in the defendant jurisdiction creates a risk of vote dilution and, conditional on such a risk, about whether the state has done enough to mitigate it by enabling minority “voting power.” (In the extreme case in which the distributions of political preferences among white and minority voters were identical, there would be no possibility of racial vote dilution. The minority group would lack distinct political interests, and minority voters’ opportunity to elect the representatives that they prefer would be unaffected by the racial makeup of electoral districts.)

As this Section explains, the four theories of racial vote dilution have quite different implications for what is a legally significant distribution of political preferences across racial groups—that is, a distribution that presents a significant risk of unlawful vote dilution—and also for what constitutes a legally sufficient opportunity to exercise voting power when such preference distributions exist. We will also see that many of the lower courts’ disagreements about how to implement the racial-polarization test are precisely the disagreements that one would expect to find if different judges subscribed, sub silentio, to different normative theories of dilution.

To be clear, this Section does not provide a definitive account of what each of the theories implies regarding legally significant preference distributions or the hallmarks of a legally sufficient state accommodation for minority voters. The theories are still inchoate, so what we say here should be taken as only suggestive. Our limited goal is to show that the theories (plausibly) point in opposite directions on a number of key questions that have befuddled and divided judges, such as the proper geographic scale of the polarization analysis, the definition of “minority candidate of choice,” the meaning of “usual defeat,” and the relative weight to be given to voting patterns in primary versus general elections.

It follows that insofar as judges exercise discretion over such matters—and currently they have a lot of discretion—inconsistent application of the polarization test should be regarded as normal and expected, rather than aberrational. Appellate judges who are worried about inconsistent (“unmanageable”) application of the polarization test ought to be at least as concerned with the normatively obscure and fragmented state of vote dilution law as they are with the absence of bright-line rules under the second and third Gingles prongs.

a) Geographic scale of the polarization inquiry.  If the three Gingles prongs were only about potential remedies, it would make sense to apply them at the scale of the plaintiff’s electoral district, irrespective of one’s normative theory of dilution. Proof that a geographic cluster of minority voters cannot elect their preferred candidates but would be able to do so in a proposed remedial district is necessary to establish that minority voting strength can be increased.

But if the Gingles test is to be diagnostic of liability, paring the universe of potential claims down to a smaller number presenting serious risks of a statutory violation, then the appropriate geographic scale will vary with one’s theory of dilution. The proportional representation and coalitional-breakdown theories require a polity-wide assessment of minority cohesion and white bloc voting, since the theories focus on the minority’s representational opportunities within the legislature as a whole.

The same may be true of the voter-discrimination theory, although this is uncertain. Pellucid in what it condemns (voter discrimination), this theory is less clear about what constitutes a legally sufficient representational opportunity for racial minorities in the teeth of voter discrimination. One might posit that an electoral system should be deemed compliant if the racial minority enjoys a realistic opportunity to elect a roughly proportional number of minority-race candidates in the jurisdiction as a whole, notwithstanding voter discrimination. Or the system might be held compliant if the number of districts where minority candidates are disadvantaged by voter discrimination is offset by an equal number where the median voter prefers minority-race candidates over otherwise-similar white candidates. Then again, one might argue that if voter discrimination is particularly pronounced in certain neighborhoods or regions of the defendant jurisdiction, compliance with § 2 should be assessed in terms of the representational opportunities provided for minority voters solely in the problematic area, rather than polity-wide.

The LULAC theory uniquely compels a district-specific polarization inquiry. It is concerned with commonality among distinct territorial communities that could form local, district-level majorities. If Asian Americans in one neighborhood are liberal Democrats and Asian Americans in another are conservative Republicans, each Asian American community could in principle claim dilution under the LULAC theory, provided of course that it could form a majority of a “natural” legislative district. By contrast, the political distance between these Asian American communities would preclude a vote dilution claim if political cohesion were assessed at the scale of the polity, as the proportional representation and coalitional-breakdown theories suggest.

Similarly, the LULAC theory calls for white bloc voting to be assessed at the level of the plaintiffs’ current legislative district, rather than throughout the jurisdiction. If a district-level white majority has radically different political preferences than the minority voters, this implies that the district was not drawn in accordance with “community of interest” districting principles. By contrast, if the local white majority’s political aims are not so discordant with those of minority voters in the district, the state’s failure to draw the majority-minority district sought by the plaintiffs is less likely to look like a grave and possibly discriminatory departure from traditional districting criteria.

b) “Legally significant” racial-group cohesion (preference polarization).  Beyond the issue of geographic scale, the theories of dilution rest on different ideas about the type of preference cohesion within and divergence between racial groups that trigger state obligations under § 2.

The sharpest contrast is between the proportional representation and voter-discrimination theories. The former requires the minority community to be strongly politically cohesive and to hang together not simply as Democrats or as Republicans but as a political faction that could be expected to form its own party in a system conducive to multiparty politics. The point of § 2 under the proportional representation theory is to provide racial groups with the opportunity to elect authentic champions for their interests. As such, if a racial minority in the defendant jurisdiction has no distinctive interests beyond the commonalities that it shares with the other members of a bridging, umbrella-like political party, then there is no occasion for judicial intervention.263

The voter-discrimination theory, by contrast, is largely indifferent to minority cohesion. If the purpose of § 2 is to remedy disparate treatment discrimination by majority-race voters, then any minority group whose members would prefer not to be discriminated against should be deemed cohesive. This minimal form of cohesion should probably be presumed as a matter of law. Notice too that under the voter-discrimination theory—unlike the proportional representation theory—there is not a monotonic relationship between polarization on the relevant dimension of political preference (candidate race) and the strength of the plaintiff’s claim. Preference polarization with respect to candidate race could reflect minority-voter discrimination against white candidates. While minority discrimination against white candidates need not foreclose a vote dilution claim under the voter-discrimination theory (what must be shown is majority-group discrimination against minority-race candidates), presumably the theory would not reward it.

It should also be clear that under the proportional representation theory, polarization on the dimension of candidate race is neither necessary nor sufficient for the racial groups to be deemed cohesive. The existence of strong political commonality within and distance between racial groups might induce voters to discriminate on the basis of candidate race; but then again, it might not, and either way it is the political commonality and not the discrimination that matters under the proportional representation theory.264

The hallmarks of a “legally significant” preference distribution are less clear under the coalitional-breakdown and LULAC theories of dilution. The LULAC theory seemingly requires the minority and white communities to be sufficiently distinct such that their inclusion in the same district amounts to a serious departure from traditional “community of interest” districting criteria.265 As for the coalitional-breakdown theory, one might posit that because the theory aims to integrate minorities into broad coalitions, a minority group should be deemed cohesive if its members have enough in common for a partisan umbrella coalition (under normal conditions) to make a pitch for the minority’s support. Then again, a pragmatic court seeking to limit the reach of § 2 might venture that the risks of enduring, race-related coalitional breakdowns are not severe unless white voters discriminate against the racial minority, or unless the minority has dramatically different political preferences from whites overall.

The competing normative theories make sense of several features of the law of group cohesion as it has developed in the lower courts. One is the conflict, noted earlier, about whether vote-share data from primary elections should receive more or less weight than data from general elections.266 Judges who subscribe to the proportional representation theory naturally look to primary elections to see whether the minority community hangs together within a political umbrella coalition, whereas judges who subscribe to the voter-discrimination or coalitional-breakdown theory need not give primaries special emphasis.

Also illuminated is the circuit split about whether white bloc voting must be “caused” by race to be legally significant (for instance, do minority candidates receive less white-voter support than otherwise-similar white candidates?).267 The voter-discrimination theory answers this question affirmatively; the proportional representation theory says that it is immaterial; and the coalitional-breakdown theory treats voter discrimination as relevant but not decisive and therefore, presumably, as a factor to be weighed at the totality-of-the-circumstances stage of a § 2 case (the position of most circuit courts).

c) Minority opportunity I: who counts as a “minority candidate of choice”?  White bloc voting per Gingles is legally significant only if it “usually” results in the defeat of minority candidates of choice.268 But who are these “candidates of [ ] choice”?269 Under the proportional representation theory, they are the ideal or nearly ideal champions of the minority community. But under the coalitional-breakdown theory, almost any candidate of the umbrella coalition preferred by the minority community should probably count as a minority candidate of choice, because the defendant’s compliance with § 2 under this theory is signaled by the minority’s integration into a competitive political coalition rather than by its ability to elect ideally preferred candidates.

The proportional representation theory supports the Third and Tenth Circuits’ practice of considering only those elections contested by an authentic champion of the minority community, as determined by the court’s “detailed, practical” evaluation of the candidate’s campaign and positions.270 For courts that regard this inquiry as judicially unmanageable, the proportional representation theory arguably suggests—in keeping with the Second Circuit—that a candidate is probably a candidate of choice only if she received a large proportion of the minority’s primary election votes.271

By contrast, the coalitional-breakdown theory attaches no particular significance to the defeat of minority champions in primary elections. Seeking to maintain influence within a coalition, a political minority might from time to time field ideally preferred candidates in primary elections. Even if the primary challenge fails, the challenge may signal that the minority faction is a force to be reckoned with and may thereby strengthen the minority group’s position within the coalition. Racially polarized voting in primaries contested by the minority’s ideal candidates no more implies that the minority is excluded from the normal push and pull of coalitional politics than polarized primary voting between Tea Partiers and mainstream Republicans signifies that Tea Partiers have inadequate sway within the Republican coalition.

The coalitional-breakdown theory lends some support to the Fourth Circuit’s rule requiring plaintiffs to produce evidence from a representative sampling of elections,272 as well as the Fourth Circuit’s definition of a “minority candidate of choice” as any candidate who won a majority of the minority votes.273 Since the theory presupposes that, in a system of plurality-winner elections, no group gets to elect its ideal candidates, there is no need for a particularized judicial inquiry into how ideal a given candidate is. A breakdown in coalitional politics can be said to have occurred, however, if data from a representative sample of elections show that the minority community almost always votes cohesively for losing candidates.274 This would suggest that there are some significant barriers to the formation of competitive, cross racial coalitions.

Consider next the voter-discrimination theory. A judge who accepts this theory needs to learn whether non-plaintiff-race voters give less support to plaintiff-race candidates than to otherwise-similar white candidates.275 It follows that any election with a minority-race candidate is at least potentially helpful for determining the legal significance of white bloc voting, regardless of whether the minority-race candidate has much backing in the minority community. The voter-discrimination theory thus lends some support to courts that have equated “minority candidate of choice” with “minority-race candidate”276 —not because minority-race candidates necessarily are, in fact, preferred by minority voters, but because the central legal question concerns the response of other voters to candidate race.277

Similarly, because the voter-discrimination theory aims to counteract discrimination on the basis of candidate race, it is at least arguable that the election of any minority-race candidates should weigh in the court’s evaluation of whether the state has made adequate accommodations for minority representation.

   What about the LULAC theory? Because this theory as applied in the namesake case focuses on objective commonalities among minority voters rather than on subjective preferences, it may not have much to say about the definition of a minority candidate of choice. But analogizing to the facts of LULAC, one might venture that a minority candidate of choice is any candidate whom most minority voters in an electoral district would prefer to the incumbent.278

d) Minority opportunity II: what constitutes “usual defeat” of minority candidates of choice?  Gingles tied the legal significance of white bloc voting to the “usual defeat” of minority-preferred candidates.279 “Usual defeat” is essentially a test of whether the state has sufficiently enabled the exercise of minority voting power. The four theories of vote dilution place quite different obligations on the state, and they therefore suggest different definitions of usual defeat—provided of course that the Gingles test is meant to be diagnostic of liability.280

Under the LULAC theory, one might say that minority-preferred candidates are “usually defeated” whenever it is substantially harder for them to get elected in the challenged district than it would be in the “natural” remedial district that the plaintiffs propose. Under the proportional representation and coalitional-breakdown theories, usual defeat must be assessed at a larger geographic scale—the polity as a whole—if the Gingles test is to separate strong from weak claims.281 Per the proportional representation theory, “usual defeat” means “defeat at such a rate as to signify that the minority lacks a fair opportunity to secure roughly proportional representation.” The coalitional-breakdown theory asks instead whether minority-preferred candidates consistently fail to win a majority of the seats in the legislative body, as the purpose of the theory is to integrate racial minorities into competitive umbrella coalitions.

What about the voter-discrimination theory? As we noted above, this theory leaves room for disagreement about what constitutes an adequate representational opportunity for minority voters in the face of white-voter discrimination.282 Perhaps proportional descriptive representation is enough, at least if the minority representatives are not stalking horses for a dominant political faction that is opposed to the minority’s agenda. Or perhaps the state must make some effort to offset white-voter discrimination, such as by creating an equal, offsetting number of districts in which the median voter is likely to discriminate in favor of minority-race candidates, or by trying to design districts so that white-voter discrimination is unlikely to determine outcomes. For present purposes, it is enough to observe that what the state owes to minority voters, given a legally significant preference distribution, is likely to be different under the voter-discrimination theory than under the proportional representation, coalitional-breakdown, or LULAC theory of vote dilution—and that the third prong of Gingles must be cashed out accordingly if the Gingles test is to be probative of liability.

3.   Summary.

The lower courts have split on a number of recurring issues in the analysis of racially polarized voting, including the relevance of monoracial elections, the weight to be given to polarization evidence from primary as opposed to general elections, the definition of the “minority candidate of choice,” the proper geographic scale of the analysis, and the question whether white bloc voting must be “caused” by race to be legally significant.283 Only the last of these issues is conventionally understood to reflect a core, normative disagreement over the meaning of racial vote dilution.284 But as this Section has shown, the competing theories of racial vote dilution have conflicting implications on each of these fronts. To the extent that governing doctrine leaves fact finders with discretion about which elections to include in the polarization analysis, how to weight them, and where to draw the line between polarized and nonpolarized preferences, inconsistent application of the polarization test should be expected—at least until such time as appellate courts make a decisive choice among the competing theories.

B.    Votes and Preferences: Of Inferential Limits, Racial Assumptions, and “the Political Stories behind the Election Returns”

Normative disagreements certainly complicate the development of clear, prescriptive rules to implement the Gingles framework, but they are not the whole story. Even if the Supreme Court were to announce that § 2 supports only one theory of racial vote dilution and then carefully explained that theory, and even if the lower courts faithfully followed that theory, it would still be extraordinarily difficult (we think impossible) for the courts to develop an objective, constraining threshold test for preference polarization based on cross tabs of candidates’ vote shares by racial group.

The underlying problem, which some judges no doubt sense but rarely wrestle with overtly, is that political actors are strategic.285 Strategic behavior muddies the relationship between racial polarization in vote shares and racial polarization in underlying political preferences.

Given strategic behavior by political actors, an analyst who wants to make inferences about latent political preferences based on candidates’ vote shares by racial group either needs lots of detailed, contextual information about the elections (what Blacksher and Menefee called “the political stories behind the election returns”286 ), or needs to make strong assumptions and throw away a lot of data. This Section develops these points.

The Gingles Court implicitly assumed that there is a strong, monotonic relationship between the average level of racial polarization in typical elections in the defendant jurisdiction and the degree of racial polarization in underlying political preferences. The same assumption undergirds the Strickland vote-share-cutoff idea: cutoffs make sense only if the average level of white crossover voting is highly informative about the intensity and pervasiveness of white discrimination against minority-race candidates or of white opposition to the minority community’s political agenda. Similarly, a number of lower courts have assumed that a candidate’s minority vote share signals how close she is to being the Platonic candidate of the minority community.287 None of these assumptions is tenable in a world of strategic political actors.

Part III.B.1 develops our point theoretically. Part III.B.2 provides illustrations using data from a survey experiment. Finally, Part III.B.3 considers how courts might best use vote-share data in light of the problems identified in the preceding sections. The main takeaway is that one of the key points of tension in the case law—between, on the one hand, the courts’ wish to avoid racial assumptions and, on the other, the courts’ desire for an objective, easy-to-apply polarization test—is essentially unavoidable, regardless of one’s theory of racial vote dilution. Moreover, even if the courts were willing to make strong assumptions, Strickland-style numeric polarization cutoffs would succeed only in targeting the most polarized communities under very unusual conditions. Specifically, the mix of observed candidate matchups in each community must be essentially identical.

A note before proceeding: Throughout this Section, we refer to “preference polarization” while remaining agnostic on the questions of what dimensions of political preference are legally relevant and what preference distributions on those dimensions rise to “legal significance” under Gingles. (As the previous Section explains, the answers to such questions depend on one’s normative theory of vote dilution.) Our premise is that per Strickland, the Supreme Court wants to foreclose vote dilution claims when preference polarization is not severe, and our aim is to show that this cannot reasonably be accomplished with bright-line cutoffs if preferences are ascertained using vote shares.

1.   Candidate attributes, strategic behavior, and polarized voting.

There is no necessary, logical relationship between racial polarization in political preferences and polarization in vote shares. Extreme preference polarization is compatible with nonpolarized voting, and minimally polarized preferences may sometimes yield extremely polarized voting.

To see this, consider a simple model in which voters have preferences over three dimensions: candidate ideology (policy positions), candidate race, and candidate quality (that is, valence traits such as probity, work effort, and intelligence). Imagine further that white and minority voters have exactly the same preferences for candidate ideology and valence traits but that most voters slightly prefer coethnic representation. In a biracial election between two candidates of similar quality who are positioned at the midpoint of the ideological spectrum, extreme racial polarization in votes is likely to occur. Because race is the only difference between the candidates, a slight preference for coethnic representation will translate into extremely polarized voting. Also, in this scenario, the difference between, say, bloc voting at the 60 percent level and at the 90 percent level is a function not of the average intensity of voters’ preference for coethnic representation but rather of the commonality of that preference.288

Now imagine another world in which there is very strong racial polarization on the ideology and race dimensions and, as before, imagine that all voters have the same preference for valence traits. White voters prefer much more conservative candidates than do minority voters, and voters of both racial groups are also willing to sacrifice a lot of candidate quality in order to elect coethnic candidates. Will we observe strong racial polarization in voting? It depends. In an election between two candidates of the same race positioned at the ideological midpoint, polarization is unlikely—the candidates are differentiated only by their valence quality, which by construction all voters value similarly. In a race between a conservative minority candidate and a liberal white candidate, polarization is also unlikely—the matchup forces voters in each group to trade off their preferences for racial and ideological proximity.289 But if a liberal minority candidate faces off against a conservative white candidate of similar quality, extreme racial polarization will probably result: minority (white) voters are closer to the minority (white) candidate on both the ideological and the race dimensions.

Now, one might respond that some of these examples are unusual, and that unusual elections will not lead to mistaken polarization inferences so long as the court averages a large number of elections or excludes outliers (as Gingles instructs).290 This response would be compelling if all elections took the form of two-candidate matchups in which the candidates’ attributes were determined by independent random draws from some pool of potential candidate attributes and if all voters had very good information about the candidates. In two-candidate, plurality-winner elections, there is no incentive for strategic voting, so votes can be interpreted as sincere expressions of preference.291 And if the candidates themselves are, in effect, just random clusters of attributes independently drawn from the pool of potential candidate attributes, then differences across jurisdictions and over time in the average level of racial bloc voting will correspond to differences in underlying political preferences.

To see the intuition here, imagine that candidates differ only on the dimension of ideology. If all elections were contests between a slightly left-of-center candidate and a slightly right-of-center candidate, then extremely polarized voting would be likely if, say, the vast majority of minority voters were slightly left of center and the vast majority of white voters were slightly right of center. The resulting inference of a big difference between white and minority voters on the ideology dimension would be mistaken.

Now imagine a counterfactual world in which each candidate’s ideology is determined by a random draw from the uniform distribution on the interval of potential ideology.292 Some elections feature two conservative candidates, other elections involve two liberal candidates, still others a centrist and a liberal, others a conservative and a liberal, and so forth, with no two candidates exactly alike. If minority voters are slightly left of center and white voters are slightly right of center, racially polarized voting will not occur in most of these elections. Specifically, white and minority voters would back the same candidate in every election between a center-left and a far-left candidate, or a center-right and a far-right candidate.

In a world of randomized candidates without strategic voting, it would also be easy enough to say how close a candidate is to being the ideal candidate of the minority community. If a given candidate runs in many elections and usually gains a strong supermajority of the minority vote, she is almost certainly close to ideal. No matter whom she is matched up against, she wins the minority vote by a landslide. Even if most candidates run only once, it would still be straightforward to estimate the minority community’s valuation of different candidate attributes and to ground candidate-of-choice determinations on estimated preferences over attributes.293

* * *

The world of randomized candidates represents an interesting thought experiment, but it bears no resemblance to actual politics. Political scientists have long realized that electoral systems create incentives for strategic behavior—by voters, candidates, parties, and other political elites.294 Observed, real-world candidates are strategically composed, as it were, by a variety of political elites seeking to advance their respective interests, including potential candidates, campaign donors, endorsers, political party insiders, and, in some jurisdictions, slating organizations.295 All of these actors operate in a strategic environment that is jointly determined by electoral institutions, by the distribution of preferences within the eligible electorate, by the distribution of civic engagement within the eligible electorate (interest in voting and attention to the campaign), and of course by the existence and behavioral tendencies of other political elites.296

Consider just a few examples of how these factors result in systematic differences across jurisdictions, districts, and types of elections in the candidate matchups that are observed, as well as in the balance between sincere and strategic voting.

Primary versus general elections.  Primary elections are likely to present much stronger incentives for strategic voting than general elections, both because of the number of serious candidates in some primary races and because many voters’ most-preferred primary candidate may be unelectable in the general election, given the distribution of preferences in the full electorate and the expected attributes of the other party’s nominee.297

Strong versus weak slating organizations.  In a world with powerful slating organizations,298 a strong candidate in each election is likely to hew to the slating organization’s positions. In a world without such gatekeepers, one may observe more diversity in candidates’ issue positions or packages of positions.

Nonpartisan versus partisan elections.  Nonpartisan elections remove one of the main centrifugal forces operating against the standard Downsian incentive for candidates to position themselves at the “ideal point” of the median voter.299 Because of this, there is likely to be less ideological distance on average between the candidates in a world of nonpartisan elections. On the other hand, there is likely to be greater heterogeneity in candidates’ positions and also a greater failure on the part of voters to see and respond to candidates’ positions.300

Instant-runoff versus plurality-winner elections.  Proponents of “instant runoff” or “rank choice” voting have persuaded a number of cities to adopt this institution, arguing that it frees voters from the dilemma of choosing between their ideal candidate and some more electable but not very exciting alternative.301 If voters behave accordingly, then first-choice votes in rank-choice-voting elections should reveal more about preference polarization than votes in ordinary plurality-winner elections.

Racist versus nonracist white voters.  In a world in which white voters are reluctant to support minority-race candidates, minority candidates are less likely to run, as well as less likely to be successful in raising money and garnering endorsements. Those minority candidates who do run are likely to be very different from “usual” minority candidates in an otherwise-identical world without white-voter discrimination.302

* * *

Further examples could be multiplied ad infinitum. But to see how these issues play out in a real-world setting, consider the facts of LULAC,303 which nicely illustrate how strategic behavior may result in the most probative of candidate matchups not occurring, as well as the risk of mistaken inferences from vote shares in the races that do occur.

At issue in LULAC was an unabashedly partisan gerrymander of Texas’s congressional districts.304 Among other things, the Republican gerrymander broke up a district with a substantial black population—but not a black majority—that the white Democrat Martin Frost had long represented.305 Black plaintiffs argued that Frost was their candidate of choice, from which it followed that Frost’s district, though not a majority-minority district, was nonetheless a black opportunity district.306 (Assume for purposes of this discussion that the court followed the proportional representation theory of vote dilution.)

Unwilling to go on vote shares alone, the courts struggled with the question whether Frost was a black candidate of choice. Frost had represented substantially the same district for more than twenty years and had never faced a black challenger.307 Local politicos offered inconsistent testimony about whether the black community really loved Frost or, if given a choice, would have preferred some other (presumably black) champion of the community.308

Why had the voters not been given that choice? One explanation is strategic abstention by would-be challengers. If Frost were the ideal or near-ideal candidate of the black community, no black politician worth his salt would bother mounting a primary challenge against him. The absence of black challengers is thus consistent with the plaintiffs’ argument that the black community was politically cohesive and that Frost was a minority candidate of choice.

But the absence of a black challenger is also consistent with the hypothesis that the black community was cohesive and that Frost was not a black candidate of choice (with “candidate of choice” defined per the proportional representation theory). Black voters were a numeric minority in Frost’s district. If they had sharply different preferences from other voters, their backing of an ideally preferred candidate in the primary—if successful—might well have resulted in a white Republican winning the general election, someone whom black voters would greatly disfavor relative to Frost. The prospect of strategic black voting for Frost could have led authentic black candidates of choice to abstain from running.

Imagine for the sake of argument that the black community in Texas was politically cohesive (within the meaning of the proportional representation theory) and that an authentic black candidate of choice had emerged to challenge Frost. Imagine further that the black community split its vote in this election. The divided black vote would count strongly against minority political cohesion if the judge followed the convention of measuring cohesion through vote shares in interracial elections, giving particular weight to primaries.309 Yet the split vote might well have been an artifact of divisions within the black community about whether to vote strategically or sincerely; of heterogeneity within the black community with respect to knowledge about the ideally preferred black candidate’s prospects in the general election; or even of heterogeneity with respect to the trade-off between representation by a senior, powerful member of Congress versus representation by a junior, less influential member whose policy positions are closer to the minority community’s. The split vote certainly would not establish that the black community lacked “common [political] beliefs, ideals, principles, [and] agendas”;310 by hypothesis, the black community was cohesive in precisely this sense.

Even when voters have no incentive to vote strategically, strategic behavior by candidates and other elites such as major campaign donors can lead to badly mistaken inferences about preference distributions. Consider a stylized example in which the elections are single stage (that is, with no separate primary), only two candidates run in each election, the racial minority is a political minority, and white voters substantially prefer white candidates to otherwise-similar minority candidates. There is a fixed pool of potential minority candidates, some of whom are good fund-raisers with strong valence qualities, and others of whom are weak. Which minority candidates will select into running? If any run, they are likely to be the weaker candidates. Strong potential candidates probably have better things to do with their time than run for office in districts where they have little chance of winning. Faced with a choice between a weak minority candidate and a strong white candidate, the minority community might split its vote, with some members voting their valence preference and others their preference for policy or for descriptive representation. This split vote could occur even if all minority voters have exactly the same ideal point on the policy dimension and the same preference for descriptive representation.

Moreover, if only weak minority candidates select into running, observed white voting patterns may suggest disparate treatment (the core question per the voter-discrimination theory) when in fact little or none occurs. Assume the data show that white voters have given less support to minority Democrats than to white Democrats—a phenomenon that courts and commentators often treat as the sine qua non of voter discrimination.311 This might reflect disparate treatment, but it could also be an artifact of the observed minority candidates’ weak valence traits, and the weakness of observed minority candidates may be completely unrelated to white prejudice. For example, if white voters tend to be conservative and most potential minority candidates are very liberal, strong minority candidates may elect not to run because they are ideologically out of step. A court that inferred disparate treatment from white voters’ lack of support for minority Democrats relative to white Democrats could be doubly in error: white voting patterns may reflect ideological as well as valence differences between minority candidates and the white candidates whom the court treats as counterfactuals.

Then again, candidates’ strategic behavior in anticipation of white-voter discrimination may lead courts to make grave errors about who is a high-quality or low-quality candidate, and in consequence to badly understate white-voter discrimination.312 Candidate quality—in the sense of innate appeal, work effort, intelligence, and so forth—is difficult to gauge, so it is natural for judges to rely on signals of quality from people with better information such as major campaign donors and endorsers. But big donors and endorsers, being strategic, are unlikely to back a candidate who has little chance of winning. If white voters do discriminate against minority-race candidates, minority candidates are unlikely to secure substantial financial support and important endorsements unless they are truly extraordinary.

If the court, trying to assess disparate treatment, were to compare white vote shares for minority candidates with a given level of financial or endorsement support to white vote shares for white candidates with similar financial or endorsement support, the court would be treating intrinsically ordinary white candidates as counterfactuals for intrinsically extraordinary minority candidates. This would bias downward the estimate of disparate treatment. On the other hand, if the court were to ignore variation in financial support, the court might end up comparing intrinsically low-quality minority candidates with high-quality white candidates, thereby overestimating the disadvantage that minority candidates suffer because of their apparent race or ethnicity.

Similar problems complicate efforts to infer polarization on the policy or ideology dimension, even if one looks only at matchups between two candidates of the same race. Strong candidates who are ideologically out of step with the district are unlikely to select into running. The true extent of polarization on the ideology dimension may be obscured by minority citizens’ failure to vote for weak candidates whose issue positions are actually closer to the minority median than the issue positions of their opponents.

Courts have long recognized that the inferences about racial polarization from any one election may be tenuous.313 Litigants are expected to present evidence from a number of elections over a period of time.314 But the problem we are discussing arises from systematic political incentives, not from random variation in candidate traits. Aggregating data from a number of elections over a period of years does not make the problem go away.

2.   Evidence from a survey experiment.

To further illustrate the limits of inference from average vote shares, we report results from experiments in which we created an electoral environment with fully randomized candidates and no incentive for strategic voting.315 Each candidate was represented with a name and a photograph, chosen so as to clearly signal the candidate’s race or ethnicity (white, black, or Latino). Respondents (“voters”) were informed about each candidate’s education, military service, and endorsements, as well as one additional piece of personal information. We established a set of possible values (“levels”) for these attributes, and we independently randomized attribute levels for each candidate. Figure 1 provides a screenshot of a hypothetical respondent’s choice task; Table 1 shows the possible levels for each attribute.

Figure 1.  Presentation of Candidate Profiles to Respondents

Table 1.  Attributes and Levels for Randomized Candidates

Attribute	Levels
Race	Black White Latino
Education	“Did not graduate from college” “Graduated from a state university” “Graduated from a state university (with honors)” “Graduated from an Ivy League university” “Graduated from an Ivy League university (with honors)”
Endorsement	“The National Association for the Advancement of Col-ored People (NAACP), the largest African American civil rights and advocacy organization in the United States” “The National Council of La Raza (NCLR), the largest Hispanic American civil rights and advocacy organization in the United States” “The Democratic Party” “The Republican Party” “The AFL-CIO, the largest association of labor unions in the United States” “The Chamber of Commerce, the largest association of business groups in the United States” “The Coalition for Sound Government”
Military Service	“Served in military; honored for valor in combat” “Served in military” “No military service”
Other Information	“Honored as whistle-blower for exposing corruption at a public agency” “Serves as board chairman of a not-for-profit hospital” “Was the starting quarterback for college football team” “Volunteers with parent-teacher association” “Charged with violating campaign-finance laws during previous run for office” “Accused of sexual harassment by several former employees”

Our randomization protocol ensured that there was no systematic relationship between a candidate’s race and any of his other characteristics or any characteristics of his opponent. Respondents had no incentive to vote strategically, as they were asked to choose between pairs of candidates in a single-stage “election.”

Though each respondent saw only eight matchups, it follows from randomization that, in expectation, the distribution of white, black, and Latino voter preferences in each election (candidate matchup) is identical.

Assume that we are in a circuit—like the Third or the Tenth—in which judges focus the polarization analysis on races contested by minority candidates who are “sponsored” by the minority community.316 In our setup, these are black candidates endorsed by the NAACP and Latino candidates endorsed by NCLR.

Matched up against “fully randomized” white candidates—candidates who could have any level for any attribute other than race or ethnicity—such black candidates of choice on average earn about 54 percent of the black vote, 35 percent of the Latino vote, and 34 percent of the white vote. Latino candidates of choice earn 57 percent of the Latino vote, 47 percent of the black vote, and 36 percent of the white vote. (These results are from a convenience sample of Amazon Mechanical Turk workers, who are younger and more liberal than the national electorate.317 The vote shares we report here would no doubt be different if we had a more representative sample.)

For courts that treat 60 percent as a rough threshold for group cohesion,318 these results suggest that the black and Latino voters in our sample are at best marginally cohesive. But these average vote shares conceal a lot of variation. Note, for example, that most of the endorsers in our study are liberal groups. This means that a “fully randomized” white candidate is on average moderately liberal. Minority-race, minority-endorsed candidates are also perceived as liberal.319 This may explain why the level of minority bloc voting in our study perhaps seems low.

How does bloc voting change if the white candidate pitted against the minority candidate of choice has a conservative endorsement? We can examine this by subsetting the data to cases in which the white candidate matched up against a black or Latino candidate of choice was endorsed by the Republican Party or the Chamber of Commerce. In these matchups, Latino voters support their candidate of choice 76 percent of the time, with white voters backing the white conservative 60 percent of the time. Black voters support their candidate of choice 57 percent of the time, with white voters favoring the white conservative 62 percent of the time. In each case, the fraction of the minority vote going to the minority candidate of choice increased; indeed, in the Latino case, the increase was a startling 19 percentage points. This is despite the fact that voter preferences did not change. The only things that changed in the two examples were the characteristics of the electoral opponent. The 19 percentage point swing in this simple example with fixed voter preferences illustrates the extreme difficulty of inferring voter preferences from simple tabulations of vote-share data—and the dubiousness of numeric bloc-voting cutoffs.

Finally, we can see the inferential problems that arise under conditions in which only weak minority candidates select into running. In our study, candidates who did not graduate from college or who had been accused of sexual harassment or campaign-finance violations were strongly disfavored by respondents in all racial and ethnic groups.320 We therefore treat the presence of one of these characteristics as indicating that the candidate is low quality. Looking at matchups featuring low-quality black, NAACP-endorsed candidates against random white candidates, we see that only 42 percent of black voters support the black candidate (down from 54 percent when the quality traits were randomized). In matchups between low-quality Latino, NCLR-endorsed candidates against random white candidates, 39 percent of Latinos support the Latino candidate (down from 57 percent when the quality traits were randomized). Again, voter preferences did not change, but vote shares for nominal candidates of choice changed dramatically.

3.   The courts’ dilemma.

Once courts face up to the mistaken assumption of Gingles—namely, that there is no monotonic relationship between racial polarization in political preferences and the average level of observed vote-share polarization in “usual” elections—what options are left? A very sensible response would be to abandon the proposition that preference polarization should be judged primarily on the basis of votes cast in actual elections. In Part IV and in other work, we discuss alternative sources of evidence.

Before turning to those alternatives, it is worth considering how courts might best use the vote-share data that have been the staple of vote dilution litigation to date. This exercise will make sense of the tensions that we observed in the case law.321 It should also prove helpful to judges and litigators insofar as they continue to rely on vote-share evidence. And it will provide a glimpse at possible futures, should the Supreme Court follow through on Strickland’s dictum and establish numeric cutoffs for “legally significant” bloc voting.

Broadly speaking, we see two ways for courts to use vote-share data in full recognition of the contingent relationship between votes and preferences: One is to adopt a subjective, informal Bayesian approach to inferring preferences from votes.322 The other is to base preference-polarization findings on a small number of “standard” elections—specifically, two-candidate, biracial, general election matchups—which are assumed as a matter of law to be comparable across jurisdictions and over time. However, neither of these approaches satisfies all the criteria that the racial-polarization test is supposed to meet.

As we have throughout this Section, we assume that the judge’s task is to make an approximate judgment about whether preference polarization in the defendant jurisdiction is severe or modest. When polarization is severe, the protections of § 2 kick in; when it is modest, minority voters must fend for themselves. (This comports with Strickland’s cutoff idea.)323

a) The informal Bayesian approach.  Judges have prior beliefs about the distribution of political preferences within white and minority communities and about the extent to which any given candidate is close to ideal for minority voters. Candidates’ vote shares by racial group also contain some information about these matters—depending on the candidates’ attributes, incentives for strategic voting, and so forth. As such, one way for judges to deal with the contingent relationship between votes and preferences is to proceed as a good Bayesian would: learn everything you can about the backstory of each election presented to the court; ask yourself, in light of that backstory, whether the candidates’ respective vote shares by racial group say something meaningful about racial polarization in preferences along any latent dimension that you believe to be legally relevant (such as candidate race, ideology, and so on); and then update your priors and repeat this process with the evidence from the next election.324

Which elections will prove most probative under the informal Bayesian approach depends on what the judge wants to learn. For example, if she wants to learn about racial polarization on the dimension of ideology, she could look for elections in which a moderate candidate faces off against an extreme candidate of the same race and of similar quality. If whites vote by large margins for very conservative candidates against centrists and if there are no significant nonideological differences between the candidates, it follows that most whites are conservative. Similarly, if minorities vote by large margins for a very liberal candidate against an otherwise-similar centrist candidate, the minority voters have revealed themselves to be cohesively liberal.

If the judge wants to investigate polarization on the dimension of candidate race, she might look for races that pit a weak white candidate against a strong minority candidate who is ideologically similar, or a strong white candidate against a weak minority candidate who is ideologically similar. If voters choose a weak coethnic candidate over a strong noncoethnic opponent, this suggests that voters care a lot about descriptive representation. But the court should not attach any significance to particular quantitative thresholds. If a very weak minority candidate running against a very strong (and ideologically similar) white candidate wins, say, 50 percent of the minority vote, that signals not a lack of cohesion but rather a strong minority preference for coethnic representation. Conversely, if 100 percent of whites vote for the white candidate in the same matchup, this says nothing about whether whites generally prefer own-race representation (or about white ideological cohesion), since the observed white vote share is readily explained by the valence gap between the candidates.

After considering all the vote-share data and reaching a conclusion about the important points of political commonality (if any) within the minority community, the informal Bayesian judge must determine which candidates were sufficiently close to the minority’s ideal as to be deemed candidates of choice. Having identified those candidates, she must then ask whether they were elected in sufficient numbers as to preclude a finding of legally significant white bloc voting. (These determinations will depend on her normative theory of dilution, not just on her updated beliefs about the distribution of political preferences in the minority and white communities.)325

The problem with the informal Bayesian approach is that it cannot be squared with the Court’s manageability objectives. It puts front and center the very thing the courts were supposed to ignore: “the political stories behind the election returns.”326 It cannot be implemented using numeric cutoffs that separate polarized from nonpolarized communities as a matter of law, or that label candidates as minority candidates of choice. Those determinations are left to hang on the judge’s prior beliefs—her stereotypes—about white and minority preferences, supplemented by whatever she may have learned from the vote-share evidence.

b) Racial assumptions “as a matter of law.”  Assume that the courts want an objective polarization test, one that can be implemented by comparing average vote shares by racial group to some Strickland-style cutoff of legal significance. Can this be done in a nonarbitrary fashion, given the contingent relationship between votes and preferences?

Per our analysis in Part III.B.1, the following conditions must be satisfied for courts to reasonably conclude that the racial groups are more politically polarized in jurisdiction A than in jurisdiction B because the average difference between observed vote shares by racial group was higher in A than in B:

1. The candidate matchups (“elections”) that go into the averages must be similar in both jurisdictions.
2. The elections must be substantially free of strategic voting.
3. The candidates in the elections must differ from one another in ways that correspond to hypothesized differences in white and minority voter preferences.
4. The candidates must be roughly equivalent in terms of attributes that are valued similarly by all voters (“valence traits”).
5. The “packages” of traits in each candidate should not run against the grain of latent preference correlations in the electorate. (A contest between a black conservative and a white liberal may not induce polarized voting if most black voters prefer black over white candidates and liberal over conservative candidates, and if most white voters prefer white over black candidates and conservative over liberal candidates.)327

These conditions suggest that an objective polarization test should be grounded on voting data from partisan, two-candidate, general elections, rather than on primaries or nonpartisan contests. Only in two-candidate general election matchups can the court be reasonably confident—without any further background information—that citizens are voting sincerely rather than strategically.328 Also, because the Democratic and Republican parties are ideological, a court relying on data from partisan general elections can be fairly confident that the two candidates are distinct from one another along policy and ideology dimensions that may matter to the voters.329

By restricting the analysis to general elections, it also becomes possible to identify and exclude elections with “against the grain” candidates (for example, black Republicans) on the basis of objective criteria. The court just needs to identify the political party that is preferred by most plaintiff-race voters and then exclude those interracial general election matchups in which the minority-race candidate was nominated by the minority-disfavored party.

Finally, though reliance on general election matchups hardly guarantees that the candidates are similar in terms of their valence attributes, the fact that the candidates have survived one screening test (the primary) means that extremely low-quality candidates are less likely to be part of the mix than if the court were using matchups from primary elections or nonpartisan races with low barriers to entry. (Note, though, that there is likely to be a strong negative correlation between incumbent valence strength and challenger valence strength. Candidates running in open-seat elections are likely to be more balanced on valence traits.)

In summary, the polarization analysis would run roughly as follows:

1. Determine the political party that is preferred by most minority voters, based on vote shares in monoracial two-candidate general election matchups.
2. Estimate vote shares by race for each candidate in interracial two-candidate general election matchups in which the minority candidate is the nominee of the minority-preferred party. Average the estimates over all elections in the analysis.
3. Make findings on minority cohesion. Cohesion might be said to exist if and only if the minority candidates in the second step received on average more than X percent of the minority vote, or if the average gap between those candidates’ minority and white vote shares exceeded some threshold of legal significance. (The latter approach would be preferable, as to some extent it controls for unmeasured differences in the quality of the minority and nonminority candidates in each matchup.)330
4. Make findings on the legal significance of white bloc voting, as appropriate. Under the Gingles plurality approach, this step reduces to asking whether the minority candidates considered at the second step were usually defeated. Under the Strickland vote-share-cutoff approach, legal significance also requires that the average crossover-voting rate not exceed some fixed level established as a matter of law. Alternatively, in those circuits where voter discrimination must be shown, a court might hold that white bloc voting is legally significant only if white candidates of the minority-preferred party received a higher proportion of the white vote than minority candidates of the minority-preferred party.

We call this the “minority-race/minority-party” protocol for quantifying racial polarization. While it avoids some of the difficulties with relying on data from primaries and nonpartisan elections, it remains deeply problematic for several reasons.

First, it severely winnows the universe of elections that may be included in the polarization analysis (even more so if the analysis is restricted to open-seat contests). If local white copartisans of the minority community generally oppose minority-race candidates, minority candidates are unlikely to win local primaries, so the polarization findings will have to be based on elections to offices responsible for much larger geographic areas (such as elections for governor, senator, or president). In some cases, particularly for Latino and Asian American plaintiffs,331 there may not be any elections in which voters in the defendant jurisdiction had the opportunity to vote for a plaintiff-race candidate who won the primary of the plaintiff group’s preferred party. In other cases, the polarization estimates will simply be noisy, affected in unknown ways by idiosyncratic features of the small number of elections that make it into the analysis. And in many cases, the estimates will be of questionable relevance. Preferences in the space of national politics may or may not be highly correlated with preferences in the space of local school board or county commission elections.

Second, the minority-race/minority-party protocol assumes that the mix of candidate matchups in each racial-polarization analysis is comparable.332 The assumption of candidate-matchup comparability would not be problematic if the analysis were restricted to data from presidential elections, because in presidential elections, voters in every jurisdiction choose between the same candidates. But relying on presidential-election data will not work for cases brought by Latino or Asian American voters; even for black vote dilution claims, we cannot be sure whether between-jurisdiction variation in racially polarized voting in the 2008 and 2012 presidential elections was due to between-jurisdiction variation in latent polarization along the race and ideology dimensions, or to idiosyncratic differences in how voters evaluated other traits of President Barack Obama, Senator John McCain, and Governor Mitt Romney (professorial mannerisms? military service? Mormonism?).333 Beyond presidential elections, the assumption of candidate-matchup comparability is extremely tenuous. There are likely to be systematic between-jurisdiction differences in partisan nominees for Congress, governor, state legislature, city council, and the like, depending on the partisan and ideological composition of the electorate. In states or districts where a party has little chance of winning general elections, high-quality candidates are less likely to seek the party’s nomination. In jurisdictions that are much more conservative or liberal than the national norm, party elites and primary voters may try to push through relatively electable candidates who deviate from the party orthodoxy in important respects. In low-profile elections, the candidates are less likely to be vetted by party elites.334 And so forth.

Moreover, as we discussed above, the white and minority candidates of a given party are likely to differ from one another in systematic ways even within a jurisdiction, due to candidate-entry incentives.335 Because of this, one cannot safely infer voter discrimination—or its absence—from the average difference in white support for white and minority candidates of the minority-preferred party (as the minority-race/minority-party protocol supposes).

The third vulnerability of the minority-race/minority-party protocol is that it effectively presumes as a matter of law that only minority-race candidates can be minority candidates of choice. This proposition has been strenuously rejected by many courts, on the ground that it reflects a “prohibited” racial assumption.336

Might there be an objective, race-neutral alternative to the minority-race/minority-party protocol? One could posit, like the Second and Ninth Circuits, that any candidate who receives strong minority support in the primary should be deemed a minority candidate of choice.337 Or one might define minority candidates of choice as those candidates who were nominated in a racially polarized primary.338 Group cohesion findings under the second and third prongs of Gingles would then be based on voting patterns in general elections that are contested by winners of racially polarized primaries (without regard to the candidate’s race).

But this race-neutral simplification could lead to very misleading conclusions. Racial polarization in the primary is only a signal of minority voters’ preference for one candidate relative to the other available choices, not a signal of how much minority voters like the preferred candidate in any absolute sense. Even as a signal of relative preference, primary polarization—or its absence—is hard to interpret because the mix of sincere and strategic voting is unknown. Lest one think this inconsequential, recall that during the 2008 primaries, most blacks did not support Obama over Clinton until he won the Iowa caucus and proved himself electable with whites.339

Bear in mind, too, that if white general election voters are hostile to the minority group and its political agenda, polarization estimates derived from general elections without a minority-race candidate may substantially understate latent preference divergence between white and minority voters.340 To put this point differently, if the court quantifies “minority cohesion” and “white bloc voting” in terms of the average general election vote share (by racial group) of all minority-preferred candidates who won racially polarized primary elections, there will be arbitrary differences across similarly polarized communities in the court-determined level of polarization. In those communities where the candidates who won polarized primaries with minority backing were often racial minorities, the observed level of general election polarization is likely to be higher, other things equal, than in communities where the candidates who won polarized primaries with minority support were usually white.

c) The upshot.  The highly contingent relationship between vote shares and political preferences has the courts in a bind. There is simply no way to concurrently satisfy the principal criteria that the Gingles test is supposed to meet. The courts could make the test objective and clear-cut—but at the price of throwing away a lot of data, implausibly assuming that the distribution of observed general election matchups is the same across jurisdictions, and making racial assumptions that many judges have strenuously resisted.

Alternatively, the courts could make the polarization inquiry expressly dependent on “the political stories behind the election returns,”341 in line with the informal Bayesian approach. This approach runs afoul of the Supreme Court’s manageability objectives. And although it avoids embedding racial assumptions in the law, it invites judges to rely on their personal stereotypes and prior beliefs about white and minority preferences.

Given these choices, the current state of the law of racially polarized voting is perhaps unsurprising. As we have seen, the courts rely heavily on vote-share data from interracial elections, but they insist—perhaps for the sake of appearances—that these are not the only relevant, probative elections. And the courts fill their opinions with vote-share estimates—creating the appearance of an objective, even mathematical inquiry—without actually limiting their discretion to interpret those numbers however they wish.342

C.    The Hidden Problem: Race-Based Assumptions and Statistics

The discussion of normative uncertainty and the contingent relationship between votes and preferences in Parts III.A and III.B implicitly assumes that it is possible to use statistical methods to estimate minority and white support for candidates running in arbitrary elections. This is true, but only if one makes an assumption that is in considerable tension with the antiessentialist strains of equal protection law.

Appendix A explains this point in detail. For readers averse to formal notation, we offer a brief, intuitive explanation here. The data in vote dilution cases typically consist of precinct-level election returns and of estimates of precinct-level demographics from the Census. The race and the vote of each individual are not observed. (Exit poll data—with individual-level observations on race and vote choice—have rarely been available.)343 The analyst uses variation across precincts in candidates’ vote shares and racial groups’ population shares to estimate the proportion of each racial group that voted for each candidate. These estimates depend on a critical assumption: the proportion of white and minority voters who support each candidate is about the same in each precinct, subject to random noise.

This is not the same as assuming that all minority voters “think alike, share the same political interests, and will prefer the same candidates at the polls”344 —the model is agnostic on exactly what proportion of minority voters supported each candidate.345 Nor does the model constrain the level of minority support for minority candidates to be the same in each precinct; random fluctuations from the estimated baseline level of support are allowed. But critically, the model does assume that support for each candidate by racial group does not vary in any systematic way across precincts. For example, minorities in relatively affluent and racially integrated precincts are treated as politically indistinguishable from minorities in poor, racially homogeneous precincts.

This assumption, which has been overlooked or even denied (perhaps inadvertently) in much of the legal academic literature,346 is not innocuous.347 Consider it in light of Kennedy’s opinion for the Court in LULAC. Kennedy wrote that a particular disputed majority-minority district should not count in the analysis of minority political opportunity, because the district combined two geographically and socioeconomically disparate clusters of Latino voters.348 The notion that these two Latino populations were jointly politically cohesive, wrote Kennedy, reflected the “prohibited assumption” of political homogeneity within ethnic groups.349 Roberts answered:

It is important to be perfectly clear about the following, out of fairness to the District Court if for no other reason: No one has made any “assumptions” about how voters in District 25 will vote based on their ethnic background. Not the District Court; not this dissent. There was a trial. At trials, assumptions and assertions give way to facts. . . . The District Court, far from “assum[ing]” that Latino voters in District 25 would “prefer the same candidate at the polls,” concluded that they were likely to do so based on statistical evidence of historic voting patterns.350

Roberts’s premise that “statistical evidence” is assumption-free is simply wrong, and the assumptions used in ecological inference are particularly strong—embodying the within-race/across-space political-homogeneity premise to which Kennedy took offense.

In principle, an expert witness could relax the assumption of spatial homogeneity within racial groups by conducting separate ecological-inference analyses on discrete geographic clusters of socioeconomically homogeneous voters. For example, the expert witness in LULAC might have fitted the ecological-inference model initially with the precincts at the northern end of District 25 (the Austin area) and then rerun the model with the precincts at the southern end of the district, along the Texas-Mexico border. Latino voters in District 25 would be deemed cohesive only if the separately fitted models showed them to have voted for the same candidates by similar margins.

However, the logic of litigation encourages pooling rather than sequential estimation. The less pooling the analyst does, the less statistically precise (maybe falsely precise) the resulting estimates are, and a court faced with imprecise estimates may well conclude that the evidence does not support a finding of minority cohesion or white bloc voting.351 Also, the separate-estimation strategy requires a method for partitioning the defendant jurisdiction into subregions within which the racial and ethnic groups are reasonably homogeneous, and it is not clear how one would do this.352

IV.  Updating Gingles

Racial vote dilution law has reached a crossroads. The Supreme Court has been persistently chipping away at this body of law, using two propositions as hammer and chisel: first, that judicial involvement in the business of adjudicating vote dilution claims requires clear rules and objective standards (at least at the screening stage);353 and second, that § 2 must be construed narrowly lest it “unnecessarily infuse race into virtually every redistricting, raising serious constitutional questions.”354

These propositions are life-threatening to the law of vote dilution as it has developed to date. Rather than being objective and rule-like, the judicial inquiry into racially polarized voting invites district courts to make the kinds of “highly political”355 and “race-based”356 judgments that the Supreme Court has warned against. Once the chasm between how the racial-polarization test was expected to work and how it actually works becomes apparent, something has to give: either the justices accept the doctrinal status quo and revise their views about judicially manageable standards and racial assumptions, or they can try to change the doctrine.

One way to change the doctrine is to convert Strickland’s dictum into law, requiring plaintiffs to show that plaintiff-race voters typically vote together at rates exceeding X percent and that other voters cross over and vote for minority-preferred candidates at rates of less than Y percent.357 But this is a fool’s errand, given the contingent relationship between preferences and votes and given the flexibility that lower courts otherwise enjoy. Lower courts could work around the thresholds of legal significance by up-weighting or discounting evidence from particular elections on the basis of the election’s supposed probativeness, or by invoking the special-circumstances doctrine. Judges who want to find for the plaintiffs in a given case would just have to dig more deeply into the stories behind the election returns to “confirm,” as it were, that the elections in which voting was most polarized are in fact the most probative and that the elections with less-polarized voting are unrevealing.

To be sure, in adopting numeric polarization cutoffs, the Court could warn against this sort of cherry-picking.358 But if the lower courts must adopt bright-line rules regarding which elections to include and exclude when analyzing polarization, the only remotely sensible choice is, as we have seen, to restrict the analysis to two-candidate, biracial, partisan, general election matchups. Yet even this solution is deeply problematic, for reasons we have canvassed.359 And it does nothing to reduce the dependence of vote dilution cases on statistical techniques that depend on the within-race/across-space political-homogeneity assumption.

The bottom line is that if the Court wants an objective screening test to filter out marginal vote dilution claims—and if the Court wants the test to be implemented without recourse to strong racial assumptions—the test should not be anchored to political preferences inferred from votes cast in actual elections.

But what is the alternative? One option is to downplay the political-preferences inquiry, perhaps shunting it to the totality-of-the-circumstances stage, and to instead screen claims using some more or less objective indicator of minority political incorporation or perhaps of intentional discrimination by the state actors responsible for the electoral system. A court following the latter approach might ask, as Judge Easterbrook has suggested, whether the number of majority-minority districts is roughly equivalent to the number likely to have been created by a race-neutral redistricting algorithm,360 or whether the state split up minority communities in violation of community of interest districting criteria.361

A court focused on minority political incorporation might ask, instead, whether there is effective partisan or party-like competition for control of the legislative body, with at least one competitive faction actively campaigning for minority votes. Or the court might simply ask whether plaintiff-race candidates have managed to get elected (perhaps excluding any plaintiff-race candidates who were actively opposed by most minority donors and endorsers). Or, more objectively: Do minority citizens vote in local elections at rates comparable to white voters?362 Some of these inquiries would require courts to dig into the political stories behind the elections, but the threshold test would not require judges to quantify the extent of preference polarization or to pronounce who is or is not a bona fide candidate of choice of the minority community.

The other path forward is for courts to continue screening claims based on preference distributions, while inviting litigants to make the associated showings using individual-level data, such as survey-based evidence of political preferences. Surveys can be used to measure racial polarization in policy preferences, general political ideology, racial attitudes, or even preferences over race and other candidate attributes as revealed by choices among randomized, hypothetical candidates. Because survey researchers are not limited to asking about vote intentions, survey-derived measures of racial polarization need not fall victim to the contingent relationship between preferences and votes. The same questions can be asked of voters in any jurisdiction, allowing meaningful between-jurisdiction and, eventually, over-time comparisons to be made.363

Back when Gingles was litigated, it would have been risible to think that racial polarization could be established through surveys. Surveys were expensive to conduct, and the available evidence from exit polls suggested that white voters often lied to survey takers about their support for black candidates. But times have changed. A number of important developments in the years since Gingles have made survey evidence much more appropriate for vote dilution litigation.

First, whites no longer overreport voting for minority candidates in opinion polls.364 This is not to deny that “social desirability” bias can contaminate responses to particularly sensitive survey questions. But researchers have developed a number of ingenious techniques for eliciting truthful answers.365 For example, respondents may be told to flip a coin, answering Question 1 if the coin comes up heads and Question 2 if it is tails. (Question 1 is innocuous, while Question 2 is very sensitive.) The respondent answers yes or no without telling the researcher which question he is answering; the researcher later backs out population-level estimates of support for the socially sensitive proposition based on the known probability of answering that question and the known level of support for the innocuous proposition.366

Another hugely important development is the emergence of low-cost, Internet-based survey methods. Over the last decade, political scientists have regularly fielded massive, cooperative surveys of the American electorate, using opt-in Internet samples that are reweighted to approximate the voting-eligible population. These surveys have been validated against a number of objective benchmarks,367 and for some purposes they can be pooled.368 Because the sample sizes are an order of magnitude larger than most previous national surveys, researchers can use the data to estimate opinions within subnational geographic units such as states or congressional districts. Also, online surveys are less affected by social-desirability biases than in-person surveys.369

New statistical techniques for modeling local public opinion as a function of demography and geography have emerged in tandem with the online surveys.370 These statistical techniques have been used to estimate public opinion within cities, counties, state legislative districts, and even school districts, using existing national data sets.371 Relatively inexpensive online surveys of the voting public within small geographic units can also be conducted by sampling from voter-registration files in proportion to estimated turnout propensities and then mailing invitations to participate in the online survey.372

Finally, statisticians and computer scientists have made enormous strides in machine learning, the science of making predictions while remaining agnostic about how best to model the event or outcome one wishes to predict.373 For VRA purposes, machine-learning tools are important not simply because they usually enable one to make better predictions than would otherwise be the case but also because they allow one to predict the political preferences of citizens with known demographic profiles (such as age, sex, education, race, income, marital status, community of residence, and so on) without making strong assumptions about political homogeneity within racial groups or even about the utility of race as a predictor of political preference. One can throw into the mix of prediction algorithms a model that assumes homogeneity within racial groups, but that model will be chosen or weighted heavily only if it does a better job predicting out-of-sample opinions than the algorithms that make no assumptions about the nature of the correlation between race and political preference.

With a set of predictions in hand, one can estimate political opinion by racial group in any geographic unit for which the US Census Bureau releases microdata. One could also estimate the “opportunity to elect” in any district by comparing the estimated preferences of the median voter in the district to the preferences of the median white and the median plaintiff-race voter in the polity as a whole (on any dimension deemed legally relevant).374

* * *

That there are plausible options for revising the vote dilution screening test—in ways that would make it more clear-cut and less dependent on racial assumptions—does not mean the revision will be easy to achieve. Two obstacles loom. One is normative dissensus over the meaning of racial vote dilution. As we have seen, it is impossible to develop a threshold test that is diagnostic of liability without a theory of dilution, and any residual discretion that district courts enjoy in implementing the threshold test is unlikely to be exercised consistently unless the Supreme Court provides normative guidance. Yet the Supreme Court has been deciding racial vote dilution cases for forty-five years without a clearly articulated theory of dilution. Mustering a five-justice majority for a theory would seem particularly difficult in the present era, in which the Court is closely divided between factions with very different approaches to race-discrimination law.375

The flexibility of the Gingles framework means that different judges can adapt it to fit their own ideas about racial vote dilution. Perhaps this suits the institutional needs of the federal judiciary, as it allows judges to decide vote dilution cases in a seemingly law-like manner without actually resolving their normative disagreements.376

The other challenge arises from the sheer range of possibilities for revising the screening test. The current convention of basing findings about minority cohesion and white bloc voting on estimates of vote shares by racial group has the significant advantage, from a manageability perspective, of strictly limiting the universe of evidentiary materials. Likewise, statistical conventions for estimating candidates’ vote shares by racial group from aggregate data have become pretty well established—even if not well justified or free of strong racial assumptions.377

If litigants were invited to make polarization showings using survey data (for example), the universe of potentially relevant data sources and statistical techniques would explode. Given this wide-open universe, it is a fair question whether the courts could develop clear-edged rules regarding data sources, modes of data analysis, and quantitative cutoffs to be used at the screening stage of vote dilution cases. The task seems better suited to administrative agencies, but Congress has not authorized any agency to issue rules with the force of law under § 2. It may be possible for the DOJ to help out by issuing advisory guidelines,378 but the Supreme Court has often been wary of administrative agencies in civil rights cases.379

In the near term, probably the best that the Supreme Court can do is to acknowledge the problems with the status quo and encourage lower courts and the DOJ to explore alternatives. That at least seems preferable to the course foretold in Strickland—for as we have seen, the establishment of numeric vote-share cutoffs for legally significant minority cohesion and white bloc voting would lock in racial vote dilution law’s dependence on within-race/across-space political-homogeneity assumptions, and the limitation on judicial discretion would prove to be either illusory or arbitrary.380

Conclusion

Voting rights law has reached a turning point. The Supreme Court recently enjoined the VRA’s preclearance regime, which required many state and local governments to obtain federal approval before implementing changes to their election laws.381 The principal remaining safeguard is the “results test” of § 2. But § 2 is also under assault, with critics—including the Supreme Court’s median justice—questioning both its manageability and its constitutionality. Whatever one makes of the constitutional question, the manageability objection has force, at least if one accepts the ideas about manageability that are traditionally invoked to justify the Gingles framework. We have shown that in its present incarnation, the test for racially polarized voting is deeply discretionary, dependent on race-based assumptions, and untethered from normative accounts of the purpose of racial vote dilution law. Now and then, judges have tried to make the polarization inquiry more rule-like, but these efforts have been largely unavailing.

The discretionary nature of the threshold test is to a large extent the ironic by-product of the courts’ decision to make candidates’ vote shares by racial group the central factor in screening vote dilution claims. This was supposed to make the test objective and constraining; but because candidate attributes mediate the relationship between racial polarization in political preferences and polarization in vote shares, and because observed candidate attributes reflect strategic choices by candidates and other actors, the screening test cannot be made constraining without also becoming absurd—that is, unless it is rebuilt on new evidentiary foundations. This Article has, we hope, made the problem clear, but the task of rebuilding the Gingles framework remains a project for another day—one that will require the sustained collaboration of courts, legal academics, statisticians, political scientists, and perhaps the DOJ.

• 42See Whitcomb v Chavis, 403 US 124, 144 (1971) (noting the possibility that a plaintiff could prove that “multi-member districts unconstitutionally operate to dilute or cancel the voting strength of racial or political elements”).
• 43Blacksher and Menefee, 34 Hastings L J at 23 (cited in note 11), quoting White v Regester, 412 US 755, 769 (1973).
• 44See Blacksher and Menefee, 34 Hastings L J at 23 & n 170 (cited in note 11) (collecting cases).
• 45Id at 43 (citations omitted).
• 46See id at 50–64.
• 47Id at 57.
• 48377 US 533 (1964).
• 49Id at 577.
• 50Blacksher and Menefee, 34 Hastings L J at 53 (cited in note 11) (emphasis added).
• 51See City of Mobile, Alabama v Bolden, 446 US 55, 70–74 (1980) (rejecting the lower courts’ reliance on a standard that allowed a plaintiff to demonstrate that a municipality violated the Fourteenth Amendment by pointing to discriminatory effects without additional evidence of discriminatory purpose).
• 52Thomas M. Boyd and Stephen J. Markman, The 1982 Amendments to the Voting Rights Act: A Legislative History, 40 Wash & Lee L Rev 1347, 1398 (1983); S Rep No 97-417 at 2 (cited in note 3) (explaining that the amendment “restores the legal standards, based on the controlling Supreme Court precedents, which applied in voting discrimination claims prior to the litigation involved in Mobile v. Bolden”).
• 5352 USC § 10301.
• 54For the legislative history of § 2, see generally Boyd and Markman, 40 Wash & Lee L Rev 1347 (cited in note 52).
• 5552 USC § 10301(b).
• 56S Rep No 97-417 at 28–29 (cited in note 3) (listing the “[t]ypical factors” that a plaintiff could show to establish a violation).
• 57Gingles, 478 US at 48–51, citing generally Blacksher and Menefee, 34 Hastings L J 1 (cited in note 11).
• 58See LULAC, 548 US at 425–26 (referring to the three Gingles factors as “threshold conditions for establishing a § 2 violation”).
• 59Gingles, 478 US at 50–51 (citation omitted).
• 60For an example of the ambiguity in Gingles, compare id at 50 (noting that the Gingles prongs are “necessary preconditions” and thereby suggesting that additional conditions must also be established), with id at 48 n 15 (noting that § 2 factors other than racial polarization as demonstrated by a satisfaction of the Gingles prongs are “supportive of, but not essential to, a minority voter’s claim”).
• 61As Brennan wrote,

  [T]he most important Senate Report factors bearing on § 2 challenges to multimember districts are the “extent to which minority group members have been elected to public office in the jurisdiction” and the “extent to which voting in the elections of the state or political subdivision is racially polarized.” . . . [O]ther factors . . . [may be] supportive of, but [are] not essential to, a minority voter’s claim.

  Id at 48 n 15. This and other passages in the Gingles plurality opinion imply that the “totality of the circumstances” referenced in the text of § 2 and enumerated in its legislative history is essentially irrelevant to vote dilution claims—except, perhaps, if it undermines the initial inference of racial polarization.

• 62See Samuel Issacharoff, Polarized Voting and the Political Process: The Transformation of Voting Rights Jurisprudence, 90 Mich L Rev 1833, 1853 & n 100 (1992); Adam B. Cox and Thomas J. Miles, Judicial Ideology and the Transformation of Voting Rights Jurisprudence, 75 U Chi L Rev 1493, 1530–35 (2008) (showing that, from 1986 to 1994, Democratic judges nearly always voted for liability conditional on the Gingles conditions being met, whereas Republican judges voted for liability conditional on Gingles in only about half of the cases during this period). For a clear and forceful articulation of the liability-standard gloss on Gingles, see Solomon v Liberty County, Florida, 899 F2d 1012, 1013 (11th Cir 1990) (en banc) (Kravitch concurring).
• 63See Gingles, 478 US at 94–100 (O’Connor concurring in the judgment) (criticizing the controlling opinion for producing a rule that “come[s] closer to an absolute requirement of proportional representation than Congress intended”).
• 64For academic accounts of Gingles’s emphasis on manageability, see, for example, Nathaniel Persily, The Law of the Census: How to Count, What to Count, Whom to Count, and Where to Count Them, 32 Cardozo L Rev 755, 780 (2011) (noting the Court’s “preference for manageable rules in this context”); Pildes, 80 NC L Rev at 1549–50 & n 93 (cited in note 36); Issacharoff, 90 Mich L Rev at 1859–71 (cited in note 62); Lani Guinier, The Triumph of Tokenism: The Voting Rights Act and the Theory of Black Electoral Success, 89 Mich L Rev 1077, 1096–1101 (1991); Pamela S. Karlan, Maps and Misreadings: The Role of Geographic Compactness in Racial Vote Dilution Litigation, 24 Harv CR–CL L Rev 173, 177–79 (1989). For examples of judicial opinions justifying Gingles on manageability grounds, see, for example, Meza v Galvin, 322 F Supp 2d 52, 57–58 (D Mass 2004) (three-judge panel) (reasoning that the Gingles conditions must have “some clarity and prescriptiveness,” lest they themselves collapse into a totality-of-the-circumstances judgment call); Rodriguez v Pataki, 308 F Supp 2d 346, 379 (SDNY 2004) (three-judge panel); Hall v Virginia, 276 F Supp 2d 528, 538 (ED Va 2003) (disallowing influence-district claims because they “would substitute the Federal Courts’ subjective estimate of minority voter influence . . . for the well established and objective rule requiring a majority-minority district”); Valdespino v Alamo Heights Independent School District, 168 F3d 848, 852 (5th Cir 1999); Hastert v State Board of Elections, 777 F Supp 634, 654 (ND Ill 1991) (three-judge panel) (“[A]n unrestricted breach of the Gingles single-district majority precondition will likely open a Pandora’s box of marginal Voting Rights Act claims by minority groups of all sizes.”); McNeil v Springfield Park District, 851 F2d 937, 942–43 (7th Cir 1988); McGhee v Granville County, North Carolina, 860 F2d 110, 116–17 (4th Cir 1988) (discussing how Gingles established reasonable bounds on potentially unbounded vote dilution jurisprudence).
• 65Issacharoff, 90 Mich L Rev at 1867–70 (cited in note 62). See also Guinier, 89 Mich L Rev at 1096 (cited in note 64).
• 66See, for example, Issacharoff, 90 Mich L Rev at 1871–73 (cited in note 62) (stating that polarized voting signifies a breakdown in normal pluralist politics); id at 1879 (stating that polarized voting signals “fundamental racial antipathy”); Guinier, 89 Mich L Rev at 1096–97 (cited in note 64) (stating that polarized voting signifies a breakdown in coalitional politics); T. Alexander Aleinikoff, The Constitution in Context: The Continuing Significance of Racism, 63 U Colo L Rev 325, 359–60 (1992) (“[R]acial bloc voting and racial appeals in the campaign process . . . provide the best evidence of private prejudice.”).
• 67See Aleinikoff, 63 U Colo L Rev at 358 (cited in note 66) (“[I]t is private prejudice . . . that is responsible for the dilution of black voting power. . . . It is the bigotry of at least a portion of the majority’s voting preferences that makes the results produced by the electoral structures unfair.”); Guinier, 89 Mich L Rev at 1102 (cited in note 64) (noting that if polarized voting is a result of prejudice, creating majority-minority districts may not solve the problem because minority officials elected to represent minority voters will be subject to discrimination in the legislative process); Issacharoff, 90 Mich L Rev at 1879 (cited in note 62) (“The persistence and extremity of the polarized voting practices in community after community, despite substantial numbers of middle-class blacks and poor whites indicates that, beyond the divergent socioeconomic interests, there must also be a more fundamental racial antipathy at work as well.”).
• 68512 US 997 (1994).
• 69Id at 1013 (noting that, “[a]s facts beyond the ambit of the three Gingles factors loom correspondingly larger, factfinders cannot rest uncritically on assumptions about the force of the Gingles factors in pointing to dilution”). The potential-remedy theory posits that unless minority voters in a defined geographic area have similar preferences to one another (the second Gingles prong), are currently unable to elect the candidates they prefer (the third Gingles prong), and would become able to elect those candidates in the remedial district(s) they propose (the first Gingles prong), their vote dilution claim is hopeless. Nothing a court could order would remedy the alleged dilution. (The theory assumes that courts can issue only single-member district remedies.) For a recent, crisp statement of the theory, see LULAC, 548 US at 495–97 (Roberts concurring in part, concurring in the judgment in part, and dissenting in part).
• 70De Grandy, 512 US at 1013–14, 1018 (concluding that such “[a]n inflexible rule would run counter to the textual command of § 2, that the presence or absence of a violation be assessed ‘based on the totality of circumstances’”).
• 71See, for example, National Association for the Advancement of Colored People v Fordice, 252 F3d 361, 373–74 (5th Cir 2001) (relying on De Grandy to reject the plaintiffs’ presumptive-liability gloss on the Gingles factors).
• 72Cox and Miles, 75 U Chi L Rev at 1537–38 & n 102 (cited in note 62).
• 73Uno v City of Holyoke, 72 F3d 973, 983 (1st Cir 1995) (“We predict that cases will be rare in which plaintiffs establish the Gingles preconditions yet fail on a section 2 claim because other facts undermine the original inference.”). We had law students code a random sample of § 2 district court opinions (see Appendix B for details), and our very rough point estimates suggest that the proportion of cases in which the district court includes language to this effect or otherwise indicates that the Gingles factors give rise to a presumption of liability is actually larger in the post–De Grandy period than in the Gingles to De Grandy period. See note 89. (Our sample size is small, however, and the difference is not statistically significant.)
• 74See generally Strickland, 556 US 1.
• 75Id at 12 (Kennedy) (plurality).
• 76Id at 14–23 (Kennedy) (plurality).
• 77Id at 40–44 (Souter dissenting).
• 78Strickland, 556 US at 23–24 (Kennedy) (plurality).
• 79See id (Kennedy) (plurality).
• 80Id at 21–22 (Kennedy) (plurality), quoting LULAC, 548 US at 446 (Kennedy) (plurality).
• 81Strickland, 556 US at 17 (Kennedy) (plurality) (emphasis added).
• 82Id (Kennedy) (plurality), quoting Holder v Hall, 512 US 874, 894 (1994) (Thomas concurring in the judgment).
• 83Strickland, 556 US at 16 (Kennedy) (plurality).
• 84Strickland Oral Argument at *7–12 (cited in note 34). See also Hall, 512 US at 904 (Thomas concurring in the judgment) (objecting that “[t]here is no set standard defining how strong the correlation [between voter race and vote choice] must be” to pass the Gingles test).
• 85Strickland Oral Argument at *7–12 (cited in note 34). See also Abrams v Johnson, 521 US 74, 92 (1997) (noting average white crossover rates of 22 to 38 percent and black crossover rates of 20 to 23 percent, and stating that this supported the district court’s finding of no “chronic bloc voting”).
• 86See Pildes, 80 NC L Rev at 1563–64 (cited in note 36) (“The political dynamics of influence are likely to be significantly different in areas where black voters and one-third of whites vote for black candidates, as compared to areas where 90% of whites consistently vote against such candidates.”).
• 87One might think that it is arbitrary and unlawful for the Supreme Court to set up any gatekeeping condition not provided for by statute. But insofar as the VRA is best understood as a common-law statute, see Christopher S. Elmendorf, Making Sense of Section 2: Of Biased Votes, Unconstitutional Elections, and Common Law Statutes, 160 U Pa L Rev 377, 448–55 (2012), the establishment of gatekeeping conditions is probably within the scope of authority delegated to the courts. See id at 411–12 (arguing that, in passing § 2, Congress intended for courts to identify “orienting norms” and to develop “strong presumptions to guide [the totality-of-the-circumstances] inquiry”).
• 88Strickland, 556 US at 21 (Kennedy) (plurality).
• 89For examples of judicial opinions emphasizing manageability, see note 64. Regarding normative diagnosis, we had law students code a stratified random sample of district court opinions for a number of attributes, including whether the court indicated that the Gingles conditions create a presumption of liability or are strongly indicative of liability—for example, by stating that “cases will be rare in which plaintiffs establish the Gingles preconditions yet fail on a section 2 claim because other facts undermine the original inference [of vote dilution].” Uno, 72 F3d at 983. For further details, see Appendix B. Interestingly, the proportion of cases in which the district court said the Gingles conditions are strongly diagnostic of liability seems to be increasing over time, from roughly 11 percent in the Gingles to De Grandy period, to about 16 percent in the De Grandy to Strickland period, to about 35 percent in the post-Strickland period. (Because our sample is small and because the coding was imperfect, these point estimates are very rough.) On the other hand, there is strong evidence that certain judges (those appointed by Democrats) behaved as if the Gingles test were tantamount to a liability standard in the Gingles to De Grandy period, but not after De Grandy. See Part II.A.
• 90Fifteen years previously, in Holder v Hall, 512 US 874 (1994), Justice Clarence Thomas urged his colleagues to abandon the field of racial vote dilution law because, as he saw it, that body of law rests on the “pernicious . . . premise” that “members of racial and ethnic groups must all think alike on important matters of public policy.” Id at 903 (Thomas concurring in the judgment). Though Gingles requires plaintiffs to demonstrate political cohesion, Thomas posited that this demand was more nominal than real: “[T]he standards we have employed for determining political cohesion have proved so insubstantial” that, in practice, the “requirement of proof of political cohesiveness . . . has proved little different from a working assumption that racial groups can be conceived of largely as political interest groups.” Id at 903–05 (Thomas concurring in the judgment). Only Justice Antonin Scalia joined Thomas’s opinion in Hall, but in the years since, a number of jurists have expressed similar concerns. See Part II.B.3.a (showing that in the lower courts, this issue often arises when determining who counts as a “minority candidate of choice,” a question that must be answered in determining whether white bloc voting is “legally significant”).
• 91See LULAC, 548 US at 433–35.
• 92The federal reports are littered with statements justifying Gingles or some gloss on Gingles as necessary to “rein[ ] in the almost unbridled discretion that section 2 gives the courts.” McNeil v Springfield Park District, 851 F2d 937, 942 (7th Cir 1988). Without such reining in, vote dilution would become “an open-ended [concept] subject to no principled means of application.” McGhee v Granville County, North Carolina, 860 F2d 110, 116 (4th Cir 1988). And for many courts, Gingles ensures that the federal courts do not end up deciding vote dilution cases on the basis of “highly political judgments.” Strickland, 556 US at 17 (Kennedy) (plurality). Similarly, legal commentators from Blacksher and Menefee to the present have understood the Gingles conditions as clear-cut factors ascertained on the basis of objective demographic data and vote counts. See note 64.
• 93See, for example, Miller v Johnson, 515 US 900, 911–12 (1995).
• 94See, for example, Johnson v California, 543 US 499, 502–03, 515 (2005) (applying strict scrutiny to California’s practice of initially segregating inmates by race during a sixty-day evaluation period, notwithstanding undisputed record evidence concerning violent prison gangs organized along racial lines).
• 95Cox and Miles, 75 U Chi L Rev at 1506–15 (cited in note 62) (hypothesizing that the rule-like Gingles prongs constrain judges more than the standard-like totality-of-the-circumstances inquiry).
• 96Id at 1534–35.
• 97Id at 1523.
• 98Id at 1525.
• 99Cox and Miles, 75 U Chi L Rev at 1511–12, 1535 (cited in note 62).
• 100See id at 1532 (showing that there was no statistically significant change in the likelihood of judges finding the Gingles factors met when they were applied in the post-1994 period).
• 101Compare id (showing in table 4 that, in the baseline specification, African American judges were 25.1 percent more likely than white judges to vote for finding the Gingles factors met conditional on applying them), with id at 1534 (showing in table 5 that, in the baseline specification, African American judges were 17.7 percent more likely than white judges to vote for finding liability under § 2 conditional on the Gingles factors being met).
• 102Compare id at 1532 (showing in table 4 that, in the baseline specification, judges with an additional African American judge on their panels were 32.1 percent more likely to vote for finding that the Gingles factors were met conditional on applying them), with id at 1534 (showing in table 5 that, in the baseline specification, judges with an additional African American judge on their panels were 24.4 percent more likely to vote for finding liability conditional on the Gingles factors being met).
• 103Though the point estimates suggest that the direct and panel effects of judge race are larger at the Gingles stage than at the totality-of-the-circumstances stage, the difference between the point estimates may not be statistically significant. It should be noted too that Cox and Miles’s analysis depends on the questionable premise that there is no intrinsic difference apart from the legal standard in the decisions judges are asked to make at the Gingles and totality-of-the-circumstances stages of a § 2 case. Cox and Miles, 75 U Chi L Rev at 1513–15 (cited in note 62). Unless one buys this equivalence idea, it does not make sense to treat differences in the correlation between judge race or ideology and judge votes at the two stages as an “effect” of the legal standard.
• 104The legal explanation developed in this paragraph is obliquely referenced in a footnote in Cox and Miles’s article. Id at 1538 n 102:

  This transformation also suggests that the Supreme Court’s decision in Johnson v De Grandy, 512 US 997 (1994), may be more consequential than is often recognized. . . . In light of our evidence, [ ] one might read the Court’s decision in De Grandy as . . . a signal to lower courts about the declining importance of the preconditions.

• 105See Part I.B.
• 106See text accompanying notes 69–72.
• 107The best casebook treatment of the subject is Samuel Issacharoff, Pamela S. Karlan, and Richard H. Pildes, The Law of Democracy: Legal Structure of the Political Process 702–05, 711–20 (Foundation 4th ed 2012).
• 108Blacksher and Menefee, 34 Hastings L J at 53 (cited in note 11).
• 109Gomez v City of Watsonville, 863 F2d 1407, 1415 (9th Cir 1988). The courts have consistently rejected efforts to rebut vote-share evidence of polarization with survey or other data on the political preferences and interests of racial communities.
• 110See, for example, id at 1416 (describing the Fifth Circuit’s rejection of survey evidence because surveys showing that African Americans and Hispanics had different political interests provided no evidence of actual voting behavior); United States v Blaine County, Montana, 363 F3d 897, 910 (9th Cir 2004) (rejecting the plaintiffs’ arguments that the district court erred in failing to consider data on low voter turnout or the lack of additional nonelectoral evidence that the plaintiffs had “distinct political concerns”). In none of the sixty-nine cases in our sample did the court determine that vote-share evidence tending to support a finding of minority cohesion or white bloc voting was outweighed or rebutted by anecdotal or survey evidence indicating a lack of cohesion.
• 111In 87 percent of the cases in our sample in which the court adjudicated a dispute over the second prong of Gingles, the coder judged the court to have relied “exclusively” or “primarily” on voting data when making findings about minority cohesion. For the third prong, the court relied exclusively or primarily on voting data in 90 percent of the cases.
• 112Gingles, 478 US at 51, 54, 57 (listing examples of special circumstances that, if combined with a minority-preferred candidate’s victory, would not provide clear or conclusive evidence against polarized voting).
• 113See id at 56 (“A showing that a significant number of minority group members usually vote for the same candidates is one way of proving the political cohesiveness necessary to a vote dilution claim.”); id at 57 (“[I]n a district where elections are shown usually to be polarized, the fact that racially polarized voting is not present in one or a few individual elections does not necessarily negate the conclusion that the district experiences legally significant bloc voting.”).

  Blacksher and Menefee did recognize that voting patterns may not be probative of underlying polarization unless one of the candidates is “publicly identified with the interests of” the minority community. Blacksher and Menefee, 34 Hastings L J at 51 (cited in note 11) (emphasis omitted). However, they believed that such candidates could be identified by courts objectively, on the basis of their minority vote share. Id at 59 (“Receipt of the racial minority’s bloc support would be the best, and necessary, indication that a candidate is identified with the minority’s particular interests.”).

• 114See, for example, Gomez, 863 F2d at 1414–17 (analyzing the “[p]olitical [c]ohesiveness of Hispanics” and then subsequently and separately analyzing racial bloc voting among majority voters). The controlling opinion in Gingles arguably suggests that both of these questions should be treated as part of the second-prong inquiry, with the third prong reserved for the question whether minority-preferred candidates are usually defeated. Gingles, 478 US at 51 (equating minority cohesion with the existence of “distinctive minority group interests”) (emphasis added). But this is only a stylistic point; nothing turns on whether white preferences are considered under the auspices of the second prong or the third.
• 115Bernard Grofman, Lisa Handley, and Richard G. Niemi, Minority Representation and the Quest for Voting Equality 83–84 (Cambridge 1992).
• 116See id. Grofman, Handley, and Niemi do note that in a small number of post-Gingles decisions, lower courts looked beyond voting patterns. Id at 67–69. However, based on their discussion of Gingles, they appear to view those decisions as misreadings of Gingles. Id at 83–84.
• 117See, for example, Barnett v City of Chicago, 141 F3d 699, 702 (7th Cir 1998) (noting that when “voting is ‘polarized,’ in the sense that members of various racial or ethnic groups have a strong preference for a candidate that belongs to their group,” courts must conduct a “more searching inquiry”) (emphasis added). In roughly 67 percent of the cases in our sample in which minority cohesion was disputed and the court reached the issue, the coder reported that the district court regarded the typical level of voting cohesion in the minority community as a matter of legal significance under the second prong. (In most of the remaining cases, it was not apparent to the coder whether the court regarded the typical level of minority voting cohesion as relevant under this prong.) By contrast, the difference between minority and white voting levels was regarded as a second-prong issue in only about 42 percent of the cases; similarly, statistical significance was treated by the court as a matter of legal significance in about 46 percent of the cases.
• 118Allan J. Lichtman and J. Gerald Hebert, A General Theory of Vote Dilution, 6 La Raza L J 1, 5 (1993).
• 119Id.
• 120Id at 6.
• 121The sole exception was African-American Voting Rights Legal Defense Fund, Inc v Missouri, 994 F Supp 1105 (ED Mo 1997), a case in which there was no evidence of racial polarization in any endogenous election. Moreover, this “exception” was arguably miscoded, because a key passage in the court’s opinion suggests that the court understood “political cohesion” within the meaning of Gingles as coalescence behind same-race candidates and against other-race candidates. Id at 1124 (“[I]t cannot be concluded from the data that African–Americans as a group have attempted to oust non-African-Americans, or that non-African-Americans as a group have attempted to oust African–Americans, from the bench.”).

  Note also that cases coded as “exclusively” using data from elections with a plaintiff-race candidate (under the second prong of Gingles) are roughly three times as common as cases coded as “primarily” using such data—that is, cases in which courts also considered data from elections without plaintiff-race candidates when ruling on minority cohesion.

• 122See note 117 and accompanying text.
• 124Id at 904 (Thomas concurring the judgment).
• 125See, for example, Shirt v Hazeltine, 336 F Supp 2d 976, 999, 1016–17 (D SD 2004) (“Factoring in the margin of error and the arbitrary nature of setting a specific point, the court holds that cohesion exists at levels above 60 percent and may exist, albeit more weakly, at lower levels.”); Johnson v Hamrick, 155 F Supp 2d 1355, 1370 (ND Ga 2001) (stating that minority bloc voting above 70 percent establishes cohesion as a matter of law, while finding cohesion based on the fact that the minority group sometimes but not always voted that cohesively).
• 126This claim is based on our review of students’ coding of the G2R.Rules.Nar and G2R.Standards.Nar fields. A codebook defining these fields is available at http://lawreview.uchicago.edu/page/elmendorf-quinn-abrajano. For the exception, see Mallory v Ohio, 38 F Supp 2d 525, 550 (SD Ohio 1997) (rejecting a vote dilution claim because the “‘average degree of racial bloc voting’ [in the instant case did] not begin to approach the 70–plus percent degree of racial bloc voting discussed by Judge Boggs in Clarke v. City of Cincinnati . . . as being typical of legally significant racial bloc voting”) (emphasis omitted). By contrast, in a number of cases in our sample, judges expressly rejected arguments for bright-line rules under the second prong of Gingles. See, for example, Fabela v City of Farmers Branch, Texas, 2012 WL 3135545, *30–34 (ND Tex) (weighing point estimates and confidence intervals in lieu of cutoffs); Large v Fremont County, Wyoming, 709 F Supp 2d 1176, 1215 (D Wyo 2010) (following the Eighth and Ninth Circuits in rejecting the 60 percent voting-cohesion requirement under the second prong of Gingles and noting that the “[d]efendants were unable to point to a single case” adopting this requirement); United States v Village of Port Chester, 704 F Supp 2d 411, 427 (SDNY 2010) (rejecting “bright-line threshold[s],” including the 60 percent rule proposed by the defendants); United States v City of Euclid, 580 F Supp 2d 584, 596 & n 15 (ND Ohio 2008) (“There are no bright line absolutes to which this Court must adhere in assessing the question of whether racial bloc voting existed in Euclid.”); Buchanan v City of Jackson, Tennessee, 683 F Supp 1515, 1527 (WD Tenn 1988) (rejecting the defendants’ proposal to hold as a matter of law that racial polarization exists if and only if the sum of the percentage of black voters voting for black candidates plus the percentage of white voters voting for white candidates exceeds 160 percent in biracial matchups).

  Reading cases beyond our sample, we found two opinions that do appear to endorse 60 percent minority cohesion as a minimum threshold for legally significant group cohesion. See Marylanders for Fair Representation, Inc v Schaefer, 849 F Supp 1022, 1056 (D Md 1994) (three-judge panel); Cane v Worcester County, Maryland, 840 F Supp 1081, 1088 n 5 (D Md 1994). See also Rodriguez v Pataki, 308 F Supp 2d 346, 388 (SDNY 2004) (three-judge panel) (noting that the plaintiffs’ expert applied the 60 percent rule in his polarization analysis).

• 127See notes 83–86 and accompanying text.
• 128See Part III.B (explaining that the observed level of polarized voting depends on candidate positions and other attributes).
• 129See Part III.B.
• 130To be clear, it is not our view that measuring minority and white cohesion in terms of preferences for coethnic candidates is necessarily misguided. Under certain conditions (for example, if voters have no other information about a candidate’s policy positions or political values), differences in the “treatment effect” of apparent candidate race on support for the candidate among groups of voters defined by race is probably a very good indicator of whether there are big intergroup differences in political preferences. But when voters know a lot about the candidates, and the candidates vary in their policy positions and other attributes, the equation of minority cohesion with “consistent voting by large margins for coethnic candidates” becomes problematic (and the problem is worse yet if there are incentives for strategic voting). See Part III.B.
• 131Gingles, 478 US at 56.
• 132We had our law-student coders assess whether the opinion in question treated various factors as intrinsically relevant to the third-prong determination (as opposed to being relevant only derivatively, insofar as the factor results in the usual defeat of minority-preferred candidates). Though “usual defeat” of minority-preferred candidates was most commonly cited as relevant (74 percent of the cases), the level of white bloc voting was deemed to be of independent legal significance in 41 percent of the cases, and evidence of disparate treatment of minority candidates by white voters (or its absence) figured into the analysis in 19 percent of the cases.
• 133See notes 83–86 and accompanying text. A few lower courts have said that if whites do not vote by very large margins against minority candidates of choice, then white bloc voting does not deny the minority community the “opportunity to elect,” because with higher minority turnout or greater minority cohesion, the minority-preferred candidates would prevail. See notes 168–71 and accompanying text.
• 134To be clear, these fifty-nine cases are the cases in our sample in which (1) the parties disputed whether the third prong of Gingles was satisfied, and (2) the district court reached this question.
• 135See, for example, Session v Perry, 298 F Supp 2d 451, 484 (ED Tex 2004) (three-judge panel) (suggesting, based on an arguable misreading of Abrams v Johnson, 521 US 74 (1997), that 30 percent white crossover voting would preclude a finding of racial polarization); Old Person v Cooney, 230 F3d 1113, 1122 (9th Cir 2000) (noting that the district court had “considered white bloc voting to be probative only if the white-preferred candidate won with a minimum of 60% of the white vote” and assuming without deciding that this was correct). See also Meza v Galvin, 322 F Supp 2d 52, 66–67 (D Mass 2004) (three-judge panel) (deeming evidence of white bloc voting “equivocal,” given that 28.1 to 42.6 percent of whites supported minority-preferred candidates); Rodriguez, 308 F Supp 2d at 388, 390 (relying on the plaintiff’s definition of “white bloc voting” as 60 percent white support for same candidate); note 126 and accompanying text (addressing vote-share cutoffs under the second prong of Gingles).
• 136For examples from our data set, see note 126 (citing cases that reject bright-line rules under the second and third prongs of Gingles). See also D. James Greiner, Re-solidifying Racial Bloc Voting: Empirics and Legal Doctrine in the Melting Pot, 86 Ind L J 447, 456 (2011) (“Lower courts mostly interpreted Justice Brennan’s opinion in Gingles as foreclosing reliance on . . . numerically defined threshold[s] or rule[s] of thumb [to delimit legally significant white bloc voting].”).
• 137See Ellen Katz, et al, Documenting Discrimination in Voting: Judicial Findings under Section 2 of the Voting Rights Act since 1982; Final Report of the Voting Rights Initiative, University of Michigan Law School, 39 U Mich J L Ref 643, 670–72 & n 138 (2006) (discussing the causation theory and collecting cases); D. James Greiner, Causal Inference in Civil Rights Litigation, 122 Harv L Rev 533, 590–97 (2008). See also Gingles, 478 US at 100 (O’Connor concurring in the judgment) (arguing in favor of a causation analysis in § 2 litigation).
• 138See Greiner, 122 Harv L Rev at 590–92 (cited in note 137). The former interpretation is probably the better reading of the causation requirement, both because it draws support from a key concurrence in Gingles and because, as Professor D. James Greiner and other methodologists have emphasized, the causal effect of voter race on vote choice is unknowable. See, for example, D. James Greiner and Donald B. Rubin, Causal Effects of Perceived Immutable Characteristics, 93 Rev Econ & Stat 775, 775–77 (2011) (explaining that while the causal effect of a “decider’s” race on a decision cannot be studied because the decider’s race cannot be manipulated, the causal effect of the decider’s perception of someone else’s race on the decision can be studied because the apparent race of a choice object is subject to manipulation).
• 139See Uno v City of Holyoke, 72 F3d 973, 981–83 (1st Cir 1995) (“[E]ven under the 1982 amendment, a lack of electoral success unrelated to race is not a proxy for a lack of opportunity to succeed.”); League of United Latin American Citizens, Council No 4434 v Clements, 999 F2d 831, 850–63 (5th Cir 1993) (en banc). See also City of Euclid, 580 F Supp 2d at 597 (explaining that evidence that white bloc voting against minority candidates increases in races between white and minority candidates offers stronger support that the plaintiffs have met the third Gingles prong); Mallory, 38 F Supp 2d at 550 (emphasizing that minority Democrats did about equally well as or even better than white Democrats); Reed v Town of Babylon, 914 F Supp 843, 877, 883 (EDNY 1996) (adopting a causation requirement and deeming evidence of white bloc voting legally insignificant because “white voters supported minority candidates slated by a political party at levels at least equal to the support enjoyed by the white candidates of that party”).
• 140In about 19 percent of the cases in our data set in which the district court ruled on the third Gingles prong, the court treated evidence of disparate treatment (or its absence) as relevant to analysis of the third prong from Gingles.
• 141See, for example, United States v Charleston County, South Carolina, 365 F3d 341, 348–49 (4th Cir 2004) (noting that although under Gingles’s third prong, “[l]egally significant white bloc voting [ ] refers to the frequency with which, and not the reason why, whites vote cohesively for candidates who are not backed by minority voters,” and that “the reason for polarized voting is a critical factor in the totality analysis”) (quotation marks omitted).
• 142Most courts have said that the defendant bears the burden of showing that the racially correlated voting patterns are not caused by race. See Greiner, 86 Ind L J at 458–60 (cited in note 136). Yet there are no established methods for defendants to make this showing.
• 143See note 112 and accompanying text.
• 144See notes 112–13 and accompanying text.
• 145See, for example, Shirt, 336 F Supp 2d at 996 (noting that interracial elections, endogenous elections—that is, “contests within the jurisdiction and for the particular office that is at issue”—and recent elections are “more probative of unequal electoral opportunity” than other elections); Ruiz v City of Santa Maria, 160 F3d 543, 552–54 (9th Cir 1998) (holding that the characteristics of an election, including the voting format and races of the candidates, influence how probative the results are and how much weight courts should give them in analyzing Gingles’s third prong). We had law-student coders summarize any “rules” or “standards” applied by district courts in the sample of opinions. While the student coders had some difficulty distinguishing between rules and standards, it appears from our review of their descriptions of the rules or standards applied by the courts that most courts made fairly subjective judgments about how much weight to give to the evidence from different elections.
• 146See, for example, Shirt, 336 F Supp 2d at 996. This was so in 90 percent of the twenty-one cases in our sample in which the district court addressed this issue.
• 147See, for example, id. This was so in 89 percent of the twenty-eight cases in our sample in which the district court addressed this issue.
• 148Curiously, this issue seems not to have been addressed in most of the cases in our sample. Student coders reported it as “not addressed” in forty-four cases and as “addressed” in only five cases. The low number may reflect coding errors, however, as the name we gave to the corresponding field in the codebook did not include the word “primary,” perhaps confusing some of the coders. Other commentators well versed in the case law regard the weight to be given to evidence from primary elections as a major issue. See, for example, Katz, et al, 39 U Mich J L Ref at 668 (cited in note 137).
• 149See id at 668–70.
• 150See, for example, Black Political Task Force v Galvin, 300 F Supp 2d 291, 305–06 (D Mass 2004) (three-judge panel) (“The relationship between race and party politics is a complicated matter and the removal of partisanship from the equation [by looking to primary elections] helps to isolate race for purposes of a vote dilution inquiry.”); Sanchez v Colorado, 97 F3d 1303, 1317 n 25 (10th Cir 1996); Bradford County NAACP v City of Starke, 712 F Supp 1523, 1534 (MD Fla 1989) (stating that primaries are “precisely the elections in which one would expect to find blacks and whites more often disagreeing on their candidate of choice”); League of United Latin American Citizens, Council No 4434, 999 F2d at 883–84; Session, 298 F Supp 2d at 478; Shirt, 336 F Supp 2d at 1008–09. See also National Association for the Advancement of Colored People, Inc v City of Niagara Falls, New York, 65 F3d 1002, 1017–19 (2d Cir 1995) (indicating that the success of the candidate backed by minority voters in a general election is not probative of the minority’s opportunity to elect if the same candidate did not garner strong minority support in the primary).
• 151See, for example, United States v Alamosa County, Colorado, 306 F Supp 2d 1016, 1033 (D Colo 2004) (“[T]he Court finds general elections to be more probative because they involve all voters in the county.”).
• 152De Grandy, 512 US at 1020. For an example of a court directly relying on this statement to downplay primary election voting patterns, see Texas v United States, 887 F Supp 2d 133, 175 (DDC 2012) (three-judge panel).
• 153See, for example, Texas, 887 F Supp 2d at 175 (“[R]equiring cohesion in the primary election distorts the role of the primary. Although minority groups sometimes coalesce around a candidate at that point in time, minority voters, like any other voters, use the primary to help develop their preferences.”); Large, 709 F Supp 2d at 1201–02 (discounting a lack of cohesion in the primary given “obvious[ ]” cohesion in the general election); Lewis v Alamance County, North Carolina, 99 F3d 600, 615 (4th Cir 1996):

  [T]he proposition that success of a minority-preferred candidate in a general election is entitled to less weight when a candidate with far greater minority support was defeated in the primary . . . is grounded in the belief that minority voters essentially take their marbles and go home whenever the candidate whom they prefer most in the primary does not prevail, a belief about minority voters that we do not share.

  (quotation marks and brackets omitted). See also Corbett v Sullivan, 202 F Supp 2d 972, 984 (ED Mo 2002) (deeming African Americans cohesive because they usually vote for Democrats, without considering how African Americans vote in Democratic primaries); Askew v City of Rome, 127 F3d 1355, 1378 (11th Cir 1997) (“Black preferred candidates [ ] are not only those who are perfectly ideologically in tune with the prevailing political sentiment in the black community. To so hold would . . . raise the possibility that none of Rome’s current black officials were truly black preferred because they are too moderate.”); Rural West Tennessee African-American Affairs Council, Inc v McWherter, 877 F Supp 1096, 1100–07 (WD Tenn 1995) (three-judge panel) (addressing the place of “influence districts” in evaluation of minority political opportunity).

• 154A judge might also reasonably decide to downplay polarization evidence from primary elections (particularly same-race primary elections) because voters tend to be less informed about primary election than general election candidates. See Christopher S. Elmendorf and David Schleicher, Informing Consent: Voter Ignorance, Political Parties, and Election Law, 2013 U Ill L Rev 363, 388–90.
• 155Blacksher and Menefee, 34 Hastings L J at 53 (cited in note 11).
• 156Black Political Task Force, 300 F Supp 2d at 309. But see Clarke v City of Cincinnati, 40 F3d 807, 813 (6th Cir 1994) (“To qualify as a ‘special’ circumstance, [ ] incumbency must play an unusually important role in the election at issue.”).
• 157Meza, 322 F Supp 2d at 67. See also Perez v Pasadena Independent School District, 958 F Supp 1196, 1222 (SD Tex 1997) (noting that when analyzing racial polarization, courts should distinguish “between serious and non-serious candidates,” a distinction “not based solely on the absolute numbers of votes received”); Cane, 840 F Supp at 1088 (ignoring an election in which, per lay testimony, the minority candidate “did not attempt to win”); Clark v Edwards, 725 F Supp 285, 297–98 (MD La 1988) (discounting evidence from an election with two low-profile black candidates).
• 158Session, 298 F Supp 2d at 484–85.
• 159See Rodriguez, 308 F Supp 2d at 392 n 59 (noting that low minority turnout in an exogenous election made the voting patterns in that election less probative); Jenkins v Red Clay Consolidated School District Board of Education, 4 F3d 1103, 1119–21 & nn 15–16 (3d Cir 1993).
• 160See Shirt, 336 F Supp 2d at 1016 (“Defendants have provided no evidence to suggest that these ballot issues touched on issues of heightened concern to the Indian community. Thus, the court need not consider this evidence.”).
• 161See Collins v City of Norfolk, Virginia, 883 F2d 1232, 1241–42 (4th Cir 1989) (discounting an election in which the mayor supported a second black candidate); Sanchez, 97 F3d at 1321–22 (discussing the possibility of discounting elections in which the alleged minority-supported candidate did not actually have minority support); Jenkins, 4 F3d at 1129 (same).
• 162Gunn v Chickasaw County, 1997 WL 33426761, *4 (ND Miss).
• 163See id; Fabela, 2012 WL 3135545 at *11–12.
• 164See Lewis, 99 F3d at 608 (“We therefore hold that . . . a district court must consider, at a minimum, a representative cross-section of elections, and not merely those in which a minority candidate appeared on the ballot.”). The First Circuit has also hinted at a representative sampling requirement, but it is unclear whether the First Circuit meant to require a sample of all elections or only of those elections contested by a minority-race candidate. See Uno, 72 F3d at 985 (stating that “all elections in the relevant time frame (or, at least, a representative sampling of them) must be studied,” but also defining “legally significant” racial polarization in terms of the defeat rate of minority candidates).
• 165See text accompanying notes 4–6, 59.
• 166This approach was used in about 78 percent of the cases in our sample of district court opinions (subsetted to cases in which the third prong of Gingles was disputed and the district court reached the issue).
• 167See, for example, Old Person, 230 F3d at 1122 (“[T]he process requires the court (1) to determine the candidate preferred by Indian voters; and (2) to determine whether whites voted as a bloc to defeat the Indian-preferred candidate.”); Jenkins v Manning, 116 F3d 685, 691 (3d Cir 1997) (describing the court’s three-part inquiry under Gingles’s second and third prongs); Sanchez, 97 F3d at 1320, quoting Collins, 883 F2d at 1237 (“‘Ascertaining whether legally significant white bloc voting exists begins with the identification of the minority members’ “preferred candidates” or “representatives of their choice.”’ . . . Only then can the district court determine whether whites vote sufficiently as a bloc usually to defeat the minority’s choice.”).
• 168This approach was used in roughly 10 percent of the cases in our sample (in which the third prong of Gingles was disputed and the district court reached the issue). By contrast, in 78 percent of the cases, the district court defined “usual defeat” or “lack of opportunity to elect” purely in terms of outcomes. (The remaining opinions could not be classified.) For examples of the opportunity-not-outcomes approach, see Jeffers v Beebe, 895 F Supp 2d 920, 930–33 (ED Ark 2012) (three-judge panel) (holding that, as a matter of law, any district in which minority citizens compose a majority of the voting-eligible population cannot violate § 2 or meet the first Gingles precondition); Perez v Perry, 835 F Supp 2d 209, 223 (WD Tex 2011) (three-judge panel) (Smith dissenting) (same); Boddie v City of Cleveland, Mississippi, 297 F Supp 2d 901, 907 (ND Miss 2004) (determining, based on white crossover-voting rates of 14 percent to 52 percent, that “minority candidates lost . . . because they failed to receive sufficient support in the majority-minority wards”); Smith v Brunswick County, Virginia, Board of Supervisors, 984 F2d 1393, 1399–1401 (4th Cir 1993) (rejecting the claim because blacks composed a majority of the voting-eligible population, notwithstanding the fact that candidates preferred by 80 percent of black voters were usually defeated); National Association for the Advancement of Colored People, Inc v City of Columbia, South Carolina, 850 F Supp 404, 420 (D SC 1993) (finding the third Gingles condition not met, notwithstanding a record of minority losses, because blacks made up a large enough share of the voting-age population that they could have elected black candidates if they had mobilized and voted cohesively).
• 169De Grandy, 512 US at 1009–16. See also id at 1014 n 11 (“[T]he ultimate right of § 2 is equality of opportunity, not a guarantee of electoral success for minority-preferred candidates of whatever race.”); Session, 298 F Supp 2d at 504 n 176 (citing De Grandy for this proposition).
• 170See, for example, Boddie, 297 F Supp 2d at 907–08 (analyzing how much support white candidates received from majority-minority wards to reject the plaintiffs’ § 2 claim under the third Gingles prong).
• 171See, for example, Jeffers, 895 F Supp 2d at 930–33.
• 172On occasion, courts have disregarded elections in which the minority candidate was backed by a local, predominantly white political faction—as opposed to being “sponsored” by the minority community. See, for example, Collins, 883 F2d at 1241–42. They have also disregarded elections in which minority turnout was unusually low. See, for example, National Association for the Advancement of Colored People, 850 F Supp at 417. But in only three cases in our sample (out of fifty-nine cases in which the court ruled on third-prong disputes) did the district court find that a minority candidate who won a plurality of the minority vote was not a candidate of choice of the minority community—and in two of those cases, the district courts’ candidate-of-choice determinations were reversed on appeal. See Jenkins, 4 F3d at 1126–28 (“According to the district court’s finding, there was a strong correlation between the race of the candidate and the preference of black voters. This raises an inference that any particular black candidate was the minority voters’ candidate of choice.”); Sanchez, 97 F3d at 1320 (reversing because, inter alia, the district court erroneously believed that “the race of the candidate is irrelevant to the racial bloc voting analysis”).
• 173See Jenkins, 4 F3d at 1129 (stating that courts should consider white candidates to be minority preferred when, among other things, they are “sponsored” by the minority community); Sanchez, 97 F3d at 1321 (same).
• 174Jenkins, 4 F3d at 1129.
• 175See note 65 and accompanying text.
• 176Ruiz, 160 F3d at 552, quoting National Association for the Advancement of Colored People, 65 F3d at 1018. See also Blaine County, 363 F3d at 910 (stating that courts should not go beyond vote shares in judging cohesion because to ask about interests is to ask an “inherently [ ] political question”).
• 177See, for example, National Association for the Advancement of Colored People, 65 F3d at 1018–19 (requiring that any candidate receive at least 50 percent of the minority vote to be considered minority preferred); Ruiz, 160 F3d at 552 (requiring that a candidate receive “sufficient votes to be elected if the election were held only among the minority group in question”—but not necessarily 50 percent of the minority vote—to qualify as minority preferred); Martinez v Bush, 234 F Supp 2d 1275, 1280 n 6 (SD Fla 2002) (three-judge panel) (adopting a standard identical to that in Ruiz v City of Santa Maria, 160 F3d 543 (9th Cir 1998)).
• 178Lewis, 99 F3d at 605–08. Thus, in the Fourth Circuit (unlike in the Second and Ninth Circuits), minority cohesion and white bloc voting can in theory be disproved with reference to white-versus-white elections in which no candidate was strongly preferred to the others by minority voters.
• 179See id at 611–15.
• 180See Ruiz, 160 F3d at 552–54.
• 181See, for example, Village of Port Chester, 704 F Supp 2d at 427–30 (sorting elections by probativeness and giving the most weight to multiracial elections when the race or ethnicity of the minority candidate was widely known); Reed, 914 F Supp at 879 (stating that exogenous elections can still be probative when they are multiracial); Clarke, 40 F3d at 810–13 (insisting that white-versus-white elections are relevant to gauging minority cohesion and white bloc voting, but criticizing the district court because it gave equal weight to evidence from interracial and monoracial elections).
• 182See Collins, 883 F2d at 1238–40 (holding that two white candidates who received minority interest group endorsements were not true candidates of choice because, inter alia, they failed to adopt certain issue positions that the court believed to be important to minority voters).
• 183See Lewis, 99 F3d at 614:

  [I]n multi-seat elections in which voters are permitted to cast as many votes as there are seats, at the very least any candidate who receives a majority of the minority vote and who finishes behind a successful candidate who was the first choice among the minority voters is automatically to be deemed a black-preferred candidate, just like the successful first choice.

• 184Id (considering borderline cases to be those in which the candidate received less than 50 percent of the minority vote).
• 185Levy v Lexington County, South Carolina, 589 F3d 708, 717–18 & n 14 (4th Cir 2009).
• 186See Harvell v Blytheville School District # 5, 71 F3d 1382, 1386 (8th Cir 1995) (holding that “[t]he preferences of the minority voters must be established on an election-specific basis, viewing all the relevant circumstances”).
• 187Clay v Board of Education of the City of St. Louis, 90 F3d 1357, 1361–62 (8th Cir 1996).
• 188Compare Askew, 127 F3d at 1377–82 (adopting a “somewhat subjective,” totality-of-the-circumstances approach to the definition of “minority-preferred candidate”), with Johnson v Hamrick, 296 F3d 1065, 1072–81 (11th Cir 2002) (sustaining a district court decision premised on the idea that any candidate who wins a majority of the minority vote is by construction a minority candidate of choice).
• 189McDaniels v Mehfoud, 702 F Supp 588, 592–93 (ED Va 1988), citing generally McGhee, 860 F2d 110.
• 190Voinovich v Quilter, 507 US 146, 153 (1993).
• 191Aldasoro v Kennerson, 922 F Supp 339, 370–72 (SD Cal 1995).
• 192Page v Bartels, 144 F Supp 2d 346, 365 (D NJ 2001) (three-judge panel).
• 193Specifically, in 90 percent of the cases, the court’s conception of “usual defeat” did not fit into one of these categories. This may be due in part to the fact that, in most cases in our sample, the court did not need to choose between these conceptions, as minority candidates of choice had almost never been elected.
• 194See, for example, Old Person v Brown, 312 F3d 1036, 1047–48 (9th Cir 2002) (holding that the racial-polarization analysis should take place at the same geographic scale as the proportionality analysis); Rural West Tennessee African-American Affairs Council v Sundquist, 209 F3d 835, 843–44 (6th Cir 2000) (same); Solomon v Liberty County Commissioners, 221 F3d 1218, 1220–21 (11th Cir 2000) (en banc) (assessing racial polarization at the polity- or countywide level); African American Voting Rights Legal Defense Fund, Inc v Villa, 54 F3d 1345, 1353–55 (8th Cir 1995) (same). In our sample of district court opinions, student coders reported that the court addressed white bloc voting at the level of the polity in forty cases, at the level of the district in twelve cases, and at an intermediate regional level in five cases.
• 195See Katz, et al, 39 U Mich J L Ref at 730–32 (cited in note 137) (noting that “De Grandy introduced ‘proportionality’ as a consideration in the totality of the circumstances analysis”).
• 196See note 194. See also, for example, Cano v Davis, 211 F Supp 2d 1208, 1239–40 (CD Cal 2002) (three-judge panel).
• 197See, for example, Solomon, 221 F3d at 1220–21 (analyzing an at-large election method for county offices at a countywide level).
• 198“Opportunity district” is a term of art for districts in which the minority community has a realistic opportunity to elect candidates of its choice. See LULAC, 548 US at 427–42.
• 199See id at 430 (“[T]he State’s creation of an opportunity district for those without a § 2 right offers no excuse for its failure to provide an opportunity district for those with a § 2 right.”).
• 200See Part III.C.
• 201This is not to say that there have never been easy cases. But the low-hanging fruit—Southern jurisdictions with at-large elections, large segregated black populations, and an unbroken history of all-white representation—has already been picked. And it is not clear to us that the racial-polarization test did any real work in those cases. They would have come out the same way if the judges applied the simpler test (as they may have sub silentio), according to which § 2 liability exists in a jurisdiction whenever: (1) no minority candidate has ever been elected; (2) there is a history of discrimination; (3) there is an at-large or multimember district electoral structure; and (4) a reasonable, compact, majority-black single-member district could be drawn.
• 202Blacksher and Menefee, 34 Hastings L J at 53 (cited in note 11). This is a requirement in those circuits that use the subjective definition of “minority candidate of choice.” See Part II.B.3.a.
• 203See Part II.B.3.a.
• 204See Part II.B.2.c.
• 205See Part II.B.2.
• 206See, for example, Clarke, 40 F3d at 813 (restricting special-circumstances arguments based on incumbency); Alamosa County, 306 F Supp 2d at 1032–33 (rejecting the defendants’ argument that three elections—one involving Hispanic candidates who downplayed their ethnicity, one involving a Hispanic Republican, and one that took place after the filing of the § 2 litigation—should be excluded or down-weighted in the analysis of white bloc voting).
• 207See Part II.B.2.a.
• 208See Part II.B.3.a.
• 209See notes 176–79 and accompanying text.
• 210See, for example, Hall v Virginia, 385 F3d 421, 423 (4th Cir 2004); McNeil, 851 F2d at 942–44; Meza, 322 F Supp 2d at 57–58; Valdespino v Alamo Heights Independent School District, 168 F3d 848, 852–53 (5th Cir 1999); Hastert v State Board of Elections, 777 F Supp 634, 654–55 (ND Ill 1991) (three-judge panel); McGhee, 860 F2d at 116. The lone federal court decision going the other way on this question is Metts v Murphy, 363 F3d 8, 11–12 (1st Cir 2004) (declining to reject a § 2 claim in which the remedial district was only 26 percent minority).
• 211Gingles, 478 US at 51 (“If the minority group is not politically cohesive, it cannot be said that the selection of a multimember electoral structure thwarts distinctive minority group interests.”).
• 212Id at 48 (“The theoretical basis for [racial vote dilution] is that where minority and majority voters consistently prefer different candidates, the majority, by virtue of its numerical superiority, will regularly defeat the choices of minority voters.”).
• 213Id at 51 (“[T]he minority must be able to demonstrate that the white majority votes sufficiently as a bloc to enable it . . . usually to defeat the minority’s preferred candidate.”).
• 214See text accompanying notes 80–91.
• 215LULAC, 548 US at 433.
• 216Gomez v City of Watsonville, 863 F2d 1407, 1415 (9th Cir 1988). See also Part II.B.1.
• 217Guinier, 108 Harv L Rev at 113 (cited in note 15).
• 218Issacharoff, Karlan, and Pildes, The Law of Democracy at 702–20 (cited in note 107) (surveying the law regarding the second and third prongs of Gingles and noting, in discussion of the third prong, that normative disputes persist in the lower courts over whether white bloc voting must be “caused” by race to be legally significant). See also Katz, et al, 39 U Mich J L Ref at 663–75 (cited in note 137) (providing an in-depth review of lower courts’ applications of the second and third prongs, without tying the different approaches to competing normative theories of vote dilution).

  The lack of scholarly attention to the ways in which normative disagreements about the meaning of racial vote dilution have colored judicial fact-finding about minority cohesion and white bloc voting is curious, because commentators in the immediate aftermath of Gingles saw the racial-polarization test as capturing something of great normative significance. See notes 65–66 and accompanying text.

• 219We had our law-student coders assess the theoretical basis of the district court opinions they coded, but after reviewing a sample of their work, we lack confidence in their judgments about this difficult question. Our assertion that “most opinions are atheoretical” is based on our own reading of the cases. In a typical case, the judge simply goes through the motions of applying Gingles and then reviewing the Senate report factors and anything else that seems relevant as part of the totality-of-the-circumstances inquiry.
• 220For a leading articulation and defense of the proportional representation theory, see Karlan, 24 Harv CR–CL L Rev at 213–19 (cited in note 64).
• 221Gingles, 478 US at 48–51.
• 222Id at 93 (O’Connor concurring in the judgment).
• 223See Part II.B.4.
• 224See De Grandy, 512 US at 1013–17 (treating “proportionality” as a key consideration in the totality-of-the-circumstances analysis); LULAC, 548 US at 436–37 (beginning the totality-of-the-circumstances analysis with an examination of proportionality).
• 225Katz, et al, 39 U Mich J L Ref at 730 (cited in note 137). Ten courts found proportionality and no liability. Four of the five courts that found a lack of proportionality found liability. No information on the remaining three decisions is reported. Id at 730–31.
• 226For articulations of the coalitional-breakdown theory of vote dilution, see Issacharoff, 90 Mich L Rev at 1860–62 (cited in note 62); Kathryn Abrams, “Raising Politics Up”: Minority Political Participation and Section 2 of the Voting Rights Act, 63 NYU L Rev 449, 452–53 (1988); Abigail M. Thernstrom, Whose Votes Count? Affirmative Action and Minority Voting Rights 23 (Harvard 1987). See also Richard H. Pildes, Foreword: The Constitutionalization of Democratic Politics, 118 Harv L Rev 29, 83–99 (2004) (raising the question whether racial vote dilution safeguards are necessary under conditions of robust two-party competition when a racial minority group is a major player in one party).
• 227See Whitcomb v Chavis, 403 US 124, 156 (1971) (rejecting the “general proposition that any group with distinctive interests must be represented in legislative halls if it is numerous enough to command at least one seat”).
• 228See Dennis C. Mueller, Public Choice III § 13.5 at 271–76 (Cambridge 2003) (discussing empirical evidence for Duverger’s law, which holds that plurality-winner elections induce a two-party system).
• 229403 US 124 (1971).
• 230412 US 755 (1973).
• 231446 US 55 (1980). See also id at 122 (Marshall dissenting) (“The vote-dilution doctrine can logically apply only to groups whose electoral discreteness and insularity allow dominant political factions to ignore them.”).
• 232See notes 51–52 and accompanying text (discussing Congress’s amendment of § 2 to overrule Bolden’s discriminatory-purpose requirement).
• 233Gingles, 478 US at 105 (O’Connor concurring in the judgment). Criticizing Brennan, O’Connor wrote that courts assessing the “voting strength” of a racial minority must consider “access to the political process generally,” not just opportunities to elect their most-preferred candidates. Id (O’Connor concurring in the judgment).
• 234539 US 461 (2003).
• 235Writing for five justices, O’Connor held that minority voting power encompasses all forms of actual or potential political influence, including the ability to elect minority candidates of choice, influence over other legislators, and opportunities for minority representatives to serve in legislative leadership positions as part of the majority coalition. Id at 478–85.
• 236LULAC, 548 US at 445–46 (Kennedy) (plurality) (“The failure to create an influence district . . . thus does not run afoul of § 2 of the Voting Rights Act.”).
• 237Strickland, 556 US at 15 (Kennedy) (plurality).
• 238See Crayton, 64 Rutgers L Rev at 982–84 (cited in note 17) (discussing the use of racial-polarization data for both offensive and defensive purposes).
• 239Strickland, 556 US at 23 (Kennedy) (plurality).
• 240Georgia, 539 US at 483.
• 241See, for example, Solomon v Liberty County, Florida, 957 F Supp 1522, 1565–71 (ND Fla 1997); National Association for the Advancement of Colored People, Inc v City of Niagara Falls, New York, 65 F3d 1002, 1019–24 (2d Cir 1995). See also Harvell v Blytheville School District # 5, 71 F3d 1382, 1395 (8th Cir 1995) (Loken dissenting) (“I am particularly distressed by the court’s . . . criticism of the manner in which successful African-American candidates ‘managed to obtain and retain their seats’ [in coalition with whites].”); United States v Alamosa County, Colorado, 306 F Supp 2d 1016, 1031 n 39 (D Colo 2004):

  [A] mix of elections, some between Hispanic candidates, some between Anglo candidates, and some with both Hispanic and Anglo candidates suggests substantial political participation by Hispanic residents and voter preferences that extend beyond ethnic identity. . . .[C]andidates must factor into their campaign strategies the practical knowledge that all voters, whether members of a minority or majority group, must seek out common political ground . . . in order to advance a political agenda.

• 242For academic articulations and defenses of the voter-discrimination theory, see Aleinikoff, 63 U Colo L Rev at 355–64 (cited in note 66); Elmendorf, 160 U Pa L Rev at 417–48 (cited in note 87).
• 243The theory may be extended to cover discrimination by conventional state actors as well. See Elmendorf, 160 U Pa L Rev at 417–48 (cited in note 87).
• 244Gingles, 478 US at 82–83 (White concurring).
• 245Id at 83 (White concurring).
• 246Id (White concurring).
• 247Id (White concurring).
• 248See Uno v City of Holyoke, 72 F3d 973, 981 (1st Cir 1995) (Selya) (“[P]laintiffs cannot prevail on a VRA § 2 claim if there is significantly probative evidence that whites voted as a bloc for reasons wholly unrelated to racial animus.”); Goosby v Town Board of the Town of Hempstead, New York, 180 F3d 476, 502–03 (2d Cir 1999) (Leval concurring) (proposing a burden-shifting standard to allow defendants to avoid § 2 liability for results unrelated to voter discrimination); League of United Latin American Citizens, Council No 4434 v Clements, 999 F2d 831, 856–59 (5th Cir 1993) (en banc) (Higginbotham) (relying on the opinions of White and O’Connor in Gingles in holding that plaintiffs cannot prevail when partisan politics rather than race best explains voting patterns); Nipper v Smith, 39 F3d 1494, 1524 (11th Cir 1994) (en banc) (Tjoflat) (“The defendant may rebut the plaintiff’s evidence by demonstrating the absence of racial bias in the voting community.”).
• 249See Katz, et al, 39 U Mich J L Ref at 670–72 & n 138 (cited in note 137) (collecting cases).
• 250See, for example, United States v Charleston County, South Carolina, 365 F3d 341, 349 (4th Cir 2004) (“[T]he reason for polarized voting is a critical factor in the totality analysis.”).
• 251The only circuit courts to have made this a necessary condition in vote dilution cases are the First and Fifth Circuits. See note 139 and accompanying text. In the sample of district court opinions coded by our students, roughly 19 percent appeared to give some weight to evidence of voter discrimination (or its absence) under the third prong of Gingles.
• 252See, for example, John Hart Ely, Democracy and Distrust: A Theory of Judicial Review 145–70 (Harvard 1980); Aleinikoff, 63 U Colo L Rev at 355–64 (cited in note 66); Bruce A. Ackerman, Beyond Carolene Products, 98 Harv L Rev 713, 732–33 (1985).
• 253For a helpful articulation of what is arguably the LULAC theory, see Richard H. Pildes, The Decline of Legally Mandated Minority Representation, 68 Ohio St L J 1139, 1142–49 (2007).
• 254Chief Justice John Roberts’s opinion nicely explains the break with previous jurisprudence. LULAC, 548 US at 493–511 (Roberts concurring in part, concurring in the judgment in part, and dissenting in part).
• 255Id at 431–35.
• 256Id at 435–38.
• 257Id at 438.
• 258De Grandy, 512 US at 1016 (reversing the district court, which had erroneously relied on the “rule of thumb” that “anything short of the maximum number of majority-minority districts consistent with the Gingles conditions would violate § 2”).
• 259LULAC, 548 US at 440.
• 260See Elmendorf, 160 U Pa L Rev at 403 (cited in note 87) (arguing that LULAC “created a building block for reorienting Section 2 toward circumstances that bespeak . . . ‘a significant likelihood’ of discriminatory intent”).
• 261See Pildes, 68 Ohio St L J at 1146–47, 1156 (cited in note 253).
• 262See Gonzalez v City of Aurora, Illinois, 535 F3d 594, 599–600 (7th Cir 2008).
• 263If the racial group essentially composes all of the party, then no showing of intraparty racial cohesion should be required. Notably, it is increasingly true that blacks essentially compose all of the Democratic Party in the Deep South. See Stephen Ansolabehere, Nathaniel Persily, and Charles Stewart III, Regional Differences in Racial Polarization in the 2012 Presidential Election: Implications for the Constitutionality of Section 5 of the Voting Rights Act, 126 Harv L Rev F 205, 211–13, 217 (2013).
• 264This presumably is why Brennan so forcefully objected to the idea that the race of successful and defeated candidates is relevant to gauging minority political opportunity. See Gingles, 478 US at 67–69 (Brennan) (plurality).
• 265See Pildes, 68 Ohio St L J at 1145 (cited in note 253).
• 266See text accompanying notes 148–54.
• 267See text accompanying notes 137–42.
• 268Gingles, 478 US at 51, 54.
• 269Id at 44.
• 270See Part II.B.3.a.
• 271See Part II.B.3.a. In Part III.B, we argue that vote shares by racial group in primary elections are not, in general, a very reliable signal of the extent to which minority voters have distinct political preferences from other members of the partisan coalition. But if primary voting is in fact racially polarized, this is some evidence of distinctiveness—although what it actually says about distinctiveness depends on the attributes of the candidates, voters’ awareness of those attributes, and the balance between sincere and strategic voting.
• 272See Parts II.B.2.c, II.B.3.a.
• 273See Part II.B.3.a.
• 274See Zoltan L. Hajnal, Who Loses in American Democracy? A Count of Votes Demonstrates the Limited Representation of African Americans, 103 Am Polit Sci Rev 37, 44–47 (2009) (showing that African Americans vote for losing candidates more often than other groups, across a wide range of elections).
• 275See Part II.B.2.b.
• 276As noted earlier, courts have almost universally treated any minority-race candidate who wins a majority of the minority vote as a candidate of choice. See note 172. And in the Fifth and Seventh Circuits, courts appear to restrict the racial-polarization analysis to elections contested by a minority-race candidate. See Campos v City of Baytown, Texas, 840 F2d 1240, 1245 (5th Cir 1988) (suggesting that the racial-polarization analysis should be based on elections in which voters have the choice of a “viable” minority candidate who was “sponsored” by the minority community); Gonzalez, 535 F3d at 598 (characterizing “candidate[ ] of [ ] choice” as a euphemism for “minority candidate”); Barnett v City of Chicago, 141 F3d 699, 702 (7th Cir 1998) (defining “polarized voting” as “members of various racial or ethnic groups hav[ing] a strong preference for a candidate that belongs to their group”).
• 277See, for example, National Association for the Advancement of Colored People, 65 F3d at 1015 (noting that “the appropriate weight to afford white-white elections in § 2 cases” is “part of a broader debate about the extent to which plaintiffs must prove racial bias to prevail under § 2”); Williams v City of Dallas, 734 F Supp 1317, 1388 (ND Tex 1990) (noting that elections with both white and minority candidates “provide the most direct test of the hypothesis that race is a factor in the election system under scrutiny”).

  That said, courts that subscribe to the voter-discrimination theory have not fully assimilated the fact that voter discrimination (in the sense of disparate treatment) cannot be estimated without a counterfactual. White bloc voting against a minority-race candidate is legally significant under this theory only if the candidate would have fared better among whites were she perceived to be white. Insofar as courts rely on observational data, the screening test under the voter-discrimination theory ought to involve a comparison of the white vote share of minority-race candidates with the white vote share of similar nonminority candidates matched up against similar opponents. The need for a counterfactual or comparison candidate is overlooked by the common convention of rebuttably presuming white-voter discrimination from a substantial divergence between white and minority vote shares for minority-preferred candidates. For examples of judicial opinions adopting this convention, see Goosby, 180 F3d at 502–03 (Leval concurring); Uno, 72 F3d at 983; Nipper, 39 F3d at 1524–26; Teague v Attala County, Mississippi, 92 F3d 283, 290 (5th Cir 1996).

• 278See LULAC, 548 US at 427 (noting that more than 90 percent of Latinos had voted against the incumbent in the previous election).
• 279Gingles, 478 US at 51.
• 280See Part I.B.
• 281See LULAC, 548 US at 436–37 (holding that in challenges to state legislative districts, proportionality should be assessed on a statewide basis).
• 282See Part III.A.2.a.
• 283See Part II.B.2.b.
• 284See note 218 and accompanying text.
• 285According to our student coders, in only three of the thirty-nine cases in our sample in which the district court passed on a dispute about the second prong of Gingles did the court overtly consider the problem of strategic behavior; similarly, in only three of the fifty-nine cases in which white bloc voting was a live issue at trial was strategic behavior factored into the court’s analysis.
• 286Blacksher and Menefee, 34 Hastings L J at 53 (cited in note 11).
• 287See Part II.B.3.a.
• 288By “within-group commonality,” we mean the proportion of voters within a racial group who share the (assumed-to-be-slight) preference for coethnic representation.
• 289Of course, somewhat polarized voting could occur in this matchup if most voters have a much stronger preference for ideological rather than racial representation, or for racial rather than ideological representation. In that case, most voters would make the same trade-off—choosing, say, the ideologically proximate candidate over the racially proximate candidate—and latent racial polarization on the ideology dimension would manifest as racial polarization in vote choice.
• 290See Parts II.B.1, II.B.2.c.
• 291This is a slight oversimplification, since some voters might vote strategically with the goal of moving the balance of power in the legislative body closer to their ideal point rather than electing a candidate at their ideal point. See Michael Tomz and Robert P. Van Houweling, Candidate Positioning and Voter Choice, 102 Am Polit Sci Rev 303, 303–04, 307–13 (2008) (estimating spatial and directional voting in the American electorate).
• 292Under the uniform distribution, every possible value is equally likely to be drawn. See, for example, Jim Pitman, Probability 28 (Springer 2006). The domain of “potential ideology” could be defined on the interval from the most liberal to the most conservative member of the electorate.
• 293For a demonstration of this point using fictional candidates, see Abrajano, Elmendorf, and Quinn, Using Experiments to Estimate Racially Polarized Voting at *30 (cited in note 41).
• 294For one of the more influential expositions of this idea, see generally Gary W. Cox, Making Votes Count: Strategic Coordination in the World’s Electoral Systems (Cambridge 1997).
• 295The literature on strategic behavior by would-be candidates and their elite supporters is vast. For some recent examples, see Kathleen Bawn, et al, A Theory of Political Parties: Groups, Policy Demands and Nominations in American Politics, 10 Persp Polit 571, 572 (2012) (arguing that political parties in the United States are “coalitions of interest groups and activists” that use the government to control policymaking); Cherie D. Maestas and Cynthia R. Rugeley, Assessing the “Experience Bonus” through Examining Strategic Entry, Candidate Quality, and Campaign Receipts in U.S. House Elections, 52 Am J Polit Sci 520, 521 (2008) (applying strategic-entry theory to nonincumbent fund-raising); Insun Kang, Richard G. Niemi, and Lynda W. Powell, Strategic Candidate Decisionmaking and Competition in Gubernatorial Nonincumbent-Party Primaries, 3 State Polit & Pol Q 353, 354 (2003) (concluding that “gubernatorial candidates respond to the same broad range of strategic considerations that influence the decisions of candidates for other high-level offices”).
• 296See Eric McGhee, et al, A Primary Cause of Partisanship? Nomination Systems and Legislator Ideology, 58 Am J Polit Sci 337, 338 (2014) (discussing the various actors in the partisan electoral system).
• 297See generally D. Sunshine Hillygus and Sarah A. Treul, Assessing Strategic Voting in the 2008 US Presidential Primaries: The Role of Electoral Context, Institutional Rules, and Negative Votes, 161 Pub Choice 517 (2014).
• 298A slating organization is a group that puts together a package of candidates who share common objectives and then campaigns on behalf of the package, encouraging voters who like one of the candidates or who share the group’s objectives to vote for all of the candidates. See generally Shanto Iyengar, Daniel H. Lowenstein, and Seth Masket, The Stealth Campaign: Experimental Studies of Slate Mail in California, 17 J L & Polit 295 (2001).
• 299Regarding the Downsian model and various forces that may lead candidates not to converge to the median voter, see generally Bernard Grofman, Downs and Two-Party Convergence, 7 Ann Rev Polit Sci 25 (2004).
• 300Greater heterogeneity is likely because the candidates do not qualify for the general election through a party-screening process. Greater failure on the part of voters to see and respond to candidates’ positions is likely because there are no party labels on the ballot, and party labels provide some information about candidate ideology. See Cheryl Boudreau, Christopher S. Elmendorf, and Scott A. MacKenzie, Informing Electorates via Election Law: An Experimental Study of Partisan Endorsements and Nonpartisan Voter Guides in Local Elections, 14 Election L J 2, 15–19 (2015) (finding that providing voters with party labels in a nonpartisan city council election strengthens spatial voting).
• 301See, for example, What Is RCV? (FairVote), archived at http://perma.cc/EXC9-62Z8 (noting that ranked-choice-voting systems “allow[ ] all voters to vote for their favorite candidate, while avoiding the fear of helping elect their least favorite candidate”).
• 302See text accompanying notes 312–13.
• 303LULAC, 548 US at 410–13 (Kennedy) (plurality).
• 304Id at 411–13 (Kennedy) (plurality).
• 305Id at 443–44 (Kennedy) (plurality).
• 306Id at 443 (Kennedy) (plurality).
• 307See Session v Perry, 298 F Supp 2d 451, 484 (ED Tex 2004) (three-judge panel).
• 308See LULAC, 548 US at 444–45 (Kennedy) (plurality).
• 309See Mallory v Ohio, 173 F3d 377, 382–84 (6th Cir 1999) (upholding a finding of no cohesion when minority vote shares varied widely across elections).
• 310League of United Latin American Citizens, Council No 4434 v Clements, 986 F2d 728, 744 (5th Cir 1993).
• 311See, for example, Charleston County, 365 F3d at 353 (emphasizing that minority-race, minority-preferred candidates were defeated more consistently than white, minority-preferred candidates); Old Person v Cooney, 230 F3d 1113, 1128 (9th Cir 2000) (emphasizing that “Indian (Indian-preferred) candidates generally received a lower percentage of white votes than did white Indian-preferred candidates in the same district”); Pildes, 80 NC L Rev at 1565–67 (cited in note 36) (questioning whether courts should regard voting as polarized if whites give similar support to black and white Democrats). To be sure, commentators and some courts have also recognized that this measure of discrimination could understate the true amount of discrimination, insofar as white support for white Democrats is colored by the perception that Democrats cater to racial minorities. See, for example, Pildes, 80 NC L Rev at 1565–67 (cited in note 36); Pamela S. Karlan and Daryl J. Levinson, Why Voting Is Different, 84 Cal L Rev 1201, 1222–24 (1996).
• 312Professor Greiner has also noted that strategic behavior by candidates (among other problems) makes it treacherous for courts to try to infer voter preferences with respect to candidate race from observational vote-share data. Greiner, 122 Harv L Rev at 596 (cited in note 137). He does not discuss the fact that strategic behavior makes it just as difficult to infer voter preferences over any other dimension from observational data.
• 313See Gingles, 478 US at 57.
• 314See id; J. Gerald Hebert, et al, The Realist’s Guide to Redistricting: Avoiding the Legal Pitfalls 55–56 (ABA 2d ed 2010).
• 315For a fuller description of the experiments and other results, see generally Abrajano, Elmendorf, and Quinn, Using Experiments to Estimate Racially Polarized Voting (cited in note 41).
• 316See Part II.B.2.a. See also notes 173–75 and accompanying text.
• 317Mechanical Turk workers are people who sign up to do small online tasks in return for (usually modest) sums of money. In recent years, this population has commonly been used as a convenience sample for research in political science and psychology. See Adam J. Berinsky, Gregory A. Huber, and Gabriel S. Lenz, Evaluating Online Labor Markets for Experimental Research: Amazon.com’s Mechanical Turk, 20 Polit Analysis 351, 366 (2012) (arguing that Amazon’s Mechanical Turk platform “provides an important way to overcome the barrier to conducting research raised by subject recruitment costs and difficulties by providing easy and inexpensive access to nonstudent adult subjects”).
• 318See notes 118–26 and accompanying text.
• 319See Abrajano, Elmendorf, and Quinn, Using Experiments to Estimate Racially Polarized Voting at *27 (cited in note 41).
• 320See id at *28.
• 321See Part II.B.
• 322The Bayesian school of statistical inference takes as given that decisionmakers have prior beliefs about the matter under investigation, and it examines how those beliefs should be updated in light of newly available data. For an introduction to Bayes’s Rule and the logic of Bayesian updating, see Simon Jackman, Bayesian Analysis for the Social Sciences xxviii–xxxiv, 1–31 (Wiley 2009).
• 323See notes 31–36 and accompanying text.
• 324For a discussion of Bayes’s Rule and judicial decisionmaking, see Richard A. Posner, How Judges Think 65–70 (Harvard 2008).
• 325See Part III.A.1.
• 326Blacksher and Menefee, 34 Hastings L J at 53 (cited in note 11).
• 327Strictly speaking, relative comparisons between jurisdictions A and B would not be distorted by the existence of some contests with “against the grain” candidates, so long as these contests occurred with similar frequency in both jurisdictions. But if there are more such contests in jurisdiction A than B, the average level of observed polarization will be lower in A even if underlying preference polarization is the same in both jurisdictions.
• 328See, for example, Hillygus and Treul, 161 Pub Choice at 529 (cited in note 297) (discussing the rate of strategic voting in the 2008 presidential election and noting that many individuals who voted strategically in the primary for Hillary Clinton voted sincerely for John McCain in the general election).
• 329It is true that basing the analysis on general election data may make it hard to determine whether the minority community is politically distinct from white copartisans—an important question under the proportional representation theory. See Part III.A.2. However, the risk of mistaken inferences from primary elections is so severe that we think primaries should be excluded from the analysis unless the court is willing to closely investigate the political stories behind the election returns. See Part III.B.1. Note also that general election results data will sometimes shed light on within-party differences, as reflected in, for example, differential rates of white and minority “defection” from various nominees of the minority-preferred party.
• 330To see the problem, imagine an election in which the Republican incumbent is very strong and only very weak challengers seek the Democratic Party’s nomination. Assume that 95 percent of whites vote for the Republican incumbent in the general election (including almost all white independents and a nontrivial number of white Democrats) and 55 percent of minority voters vote for the minority-race Democratic nominee. The “low” rate of minority voting for the coethnic candidate might be thought to suggest noncohesion, but in this example it occurs due to the weakness of the minority candidate and the strength of his opponent. The gap (50 percentage points) between the white and minority vote shares for the minority candidate is more telling of whether the minority group has distinct political interests. A few courts have recognized this point.
• 331See Andrew L. Aoki and Okiyoshi Takeda, Asian American Politics 111 (Polity 2008) (describing limited Asian American success in primary elections).
• 332Only if the underlying candidate matchups are the same or very similar can a court conclude that jurisdiction A is more polarized than B because the average difference between white and minority vote shares for minority-race candidates of the minority-preferred party is higher in A than in B.
• 333Also, with just a single matchup between a center-left black candidate and a center-right white candidate, one cannot learn much about the distribution of voter preferences within racial groups. Extremely polarized voting will be observed in both (1) a community in which white voters are slightly right of center and slightly prefer white to black candidates and black voters are slightly left of center and slightly prefer black to white candidates, and (2) a community in which white voters are extremely conservative and hate black candidates and black voters are extremely liberal and strongly prefer black to white candidates.
• 334See Casey Byrne Knudsen Dominguez, Before the Primary: Party Participation in Congressional Nominating Processes *125 (unpublished PhD dissertation, UC Berkeley, Spring 2005), archived at http://perma.cc/Q2XJ-2YPY (offering “evidence that although the party network does, naturally, support higher rather than lower quality candidates, it is more likely to do so in unified ways in races that are important to the party”).
• 335See text accompanying notes 311–13.
• 336See Part II.B.3.a. Perhaps the solution here is for courts to expressly treat “candidate of choice” as a legal fiction, derived from courts’ senses of what the VRA should be interpreted to achieve. If “candidate of choice” were recognized as a legal fiction, then a court could say, for instance, that only minority-race candidates “count” in the opportunity-to-elect analysis, not because they are the only candidates who could be bona fide candidates of choice of the minority community but because the purpose of vote dilution law is to counteract white-voter discrimination against minority-race candidates.
• 337See National Association for the Advancement of Colored People, 65 F3d at 1019 (noting that “the ballot box provides the best and most objective proxy for determining who constitutes a representative of choice” but cautioning that “a court need not treat the candidate [receiving greater than 50 percent minority support in the general election] as minority-preferred when another candidate receiving greater support in the primary failed to reach the general election”); Ruiz v City of Santa Maria, 160 F3d 543, 552 (9th Cir 1998) (following National Association for the Advancement of Colored People, Inc v City of Niagara Falls, New York, 65 F3d 1002 (2d Cir 1995)).
• 338This would likely be an improvement over the Second and Ninth Circuits’ approach, since some candidates may receive strong minority support in the primary because of valence traits rather than because they cater to the minority’s particular interests or concerns.
• 339See Paul Steinhauser, Poll: Obama Makes Big Gains among Black Voters (CNN, Jan 19, 2008), archived at http://perma.cc/X8G9-GVAQ.
• 340The voting patterns in elections without a minority-race candidate obviously will not reflect voters’ preferences with respect to candidate race, and they may not reflect voters’ preferences with respect to the political agenda of minority elites. Even if one of the candidates in a white-versus-white election shares the minority elite’s agenda, this may not be apparent to many white voters, particularly in low-profile political races.
• 341Blacksher and Menefee, 34 Hastings L J at 53 (cited in note 11).
• 342One other implication of our analysis is worth noting. To the extent that courts evaluating proposed remedial districts (at the remedy stage of a vote dilution case) aim to gauge whether the district will “perform” for minority voters, the courts should not simply assume that minority candidates in the remedial districts will receive the same shares of white and minority votes that minority candidates typically received prior to the remedy. The reason is that different types of candidates—perhaps stronger minority candidates—will select into running in the remedial district than they will under the system of districts (or at-large elections) that the court found unlawful. Thanks to Professor Nathaniel Persily for suggesting this point.
• 343See Bernard Grofman, Expert Witness Testimony and the Evolution of Voting Rights Case Law, in Bernard Grofman and Chandler Davidson, eds, Controversies in Minority Voting: The Voting Rights Act in Perspective 197, 217 (Brookings 1992).
• 344LULAC, 548 US at 433, quoting Miller v Johnson, 515 US 900, 920 (1995).
• 345That is, the model allows the average proportion of voters in a racial group who support each candidate group to vary from 0 to 1. This is the sense in which it does not assume that all voters of the group think alike.
• 346In their influential book Minority Representation and the Quest for Voting Equality, Grofman, Handley, and Niemi acknowledge that “homogeneous precinct analysis” (estimating minority or white vote shares as the average vote share in racially homogeneous precincts) may yield misleading inferences because of systematic differences in political preferences between voters of the same race who live in homogeneous as opposed to integrated neighborhoods. Grofman, Handley, and Niemi, Minority Representation at 85 (cited in note 115). But they state that the “plausibility of th[e] assumption” that homogeneous-precinct voters are indistinguishable from voters of the same race in heterogeneous precincts can be “check[ed] . . . by comparing results from a homogeneous precinct analysis with those from an ecological regression (which, of course, includes data from racially mixed precincts).” Id. They fail to note that ecological regression estimates depend on the very same homogeneity assumption. Grofman, Handley, and Niemi’s elision is repeated in the leading casebook on election law and the VRA. See Issacharoff, Karlan, and Pildes, The Law of Democracy at 678 (cited in note 107). See also Crayton, 64 Rutgers L Rev at 981 (cited in note 17) (acknowledging the problematic assumption behind homogeneous-precinct analysis and presenting ecological inference as “the best available approach”).
• 347In Garza v County of Los Angeles, California, 756 F Supp 1298, 1333–34 (CD Cal 1990), the defendants’ experts pointed to the constancy assumption in trying to impugn the plaintiffs’ experts’ reliance on ecological regression. The defense showed that a different modeling assumption—the assumption that the probability of supporting a given candidate is constant across all voters within a precinct (while varying between precincts)—can generate drastically different estimates of racial polarization. The court found that this alternative assumption “impede[d] [the defendants’ model] from detecting the presence of polarized voting” and as such that it was “not a reliable method of inferring group voting behavior.” Id at 1334. This misses the point: nothing in the data allows one to adjudicate between these modeling assumptions. It would be just as apt to say that the standard, within-race homogeneity assumption impeded the plaintiff’s model from detecting the presence of within-precinct political homogeneity across racial groups.
• 348LULAC, 548 US at 433–34.
• 349Id at 433.
• 350Id at 500 (Roberts concurring in part, concurring in the judgment in part, and dissenting in part) (emphasis added).
• 351See Grofman, Handley, and Niemi, Minority Representation at 83–84 (cited in note 115) (interpreting Gingles as requiring evidence of a statistically significant correlation between voter race and vote choice at the second prong).
• 352We tried to obtain the experts’ reports to the district court in LULAC in order to figure out the geographic region over which they pooled, but the reports are not available free of charge. The portions of the record provided on appeal to the Supreme Court do not address the pooling issue.
• 353See Strickland, 556 US at 17–20 (Kennedy) (plurality) (adopting the “majority-minority requirement” for § 2 claims because of the need for “workable standards”); Hall, 512 US at 881 (ruling out § 2 challenges to the size of a legislative body for want of an “objective and workable standard for choosing a reasonable benchmark by which to evaluate a challenged voting practice”); LULAC, 548 US at 485–86 (Souter concurring in part and dissenting in part) (acknowledging the need for “clear-edged rule[s]” to govern the threshold stage of vote dilution cases).
• 354LULAC, 548 US at 445–46 (Kennedy) (plurality). See also Strickland, 556 US at 21–22 (Kennedy) (plurality).
• 355Strickland, 556 US at 17 (Kennedy) (plurality).
• 356Id at 18 (Kennedy) (plurality).
• 357See notes 83–86 and accompanying text.
• 358Compare the Fourth Circuit’s “representative sample” requirement. See note 178 and accompanying text.
• 359See Part III.B.3.
• 360See Gonzalez v City of Aurora, Illinois, 535 F3d 594, 599–600 (7th Cir 2008).
• 361Professor Nicholas O. Stephanopoulos’s spatial-diversity index provides one way of quantifying whether an actual or proposed majority-minority district should be regarded as “natural”—that is, drawn in accordance with community of interest principles. Nicholas O. Stephanopoulos, Spatial Diversity, 125 Harv L Rev 1903, 1912–17 (2012).
• 362Note, however, that estimates of minority-voter participation in an election will often depend on the same ecological-inference techniques (with their hidden racial assumptions) used to generate vote-share estimates.
• 363We are currently working on several projects to develop survey-based estimates of group political cohesion and voter discrimination, but owing to space limitations we cannot get into the details here. See note 41.
• 364See Daniel J. Hopkins, No More Wilder Effect, Never a Whitman Effect: When and Why Polls Mislead about Black and Female Candidates, 71 J Polit 769, 773–76 (2009).
• 365These include so-called list experiments, endorsement experiments, and randomized-response surveys. For examples of leading recent works, see generally Bryn Rosenfeld, Kosuke Imai, and Jacob N. Shapiro, An Empirical Validation Study of Popular Survey Methodologies for Sensitive Questions, Am J Polit Sci (forthcoming), archived at http://perma.cc/MAS7-8JZ3; Graeme Blair, Kosuke Imai, and Yang-Yang Zhou, Design and Analysis of the Randomized Response Techniques, 110 J Am Stat Assn 1304 (2015); Graeme Blair, Kosuke Imai, and Jason Lyall, Comparing and Combining List and Endorsement Experiments: Evidence from Afghanistan, 58 Am J Polit Sci 1043 (2014); Graeme Blair and Kosuke Imai, Statistical Analysis of List Experiments, 20 Polit Analysis 47 (2012).
• 366See Blair, Imai, and Zhou, 110 J Am Stat Assn at 1307–15 (cited in note 365).
• 367See, for example, Stephen Ansolabehere and Bryan F. Schaffner, Does Survey Mode Still Matter? Findings from a 2010 Multi-mode Comparison, 22 Polit Analysis 285, 286 (2014); Stephen Ansolabehere and Douglas Rivers, Cooperative Survey Research, 16 Ann Rev Polit Sci 307, 311 (2013).
• 368See, for example, Chris Tausanovitch and Christopher Warshaw, Measuring Constituent Policy Preferences in Congress, State Legislatures, and Cities, 75 J Polit 330, 333–34 (2013) (describing the pooling process for large-scale surveys).
• 369See Frauke Kreuter, Stanley Presser, and Roger Tourangeau, Social Desirability Bias in CATI, IVR, and Web Surveys: The Effects of Mode and Question Sensitivity, 72 Pub Opinion Q 847, 858 (2008).
• 370See, for example, Elmendorf and Spencer, 115 Colum L Rev at 2204–14 (cited in note 40) (using multilevel regression with poststratification); Christopher S. Elmendorf and Douglas M. Spencer, The Geography of Racial Stereotyping: Evidence and Implications for VRA Preclearance after Shelby County, 102 Cal L Rev 1123, 1156–59 (2014) (using disaggregation and multilevel regression); Yair Ghitza and Andrew Gelman, Deep Interactions with MRP: Election Turnout and Voting Patterns among Small Electoral Subgroups, 57 Am J Polit Sci 762, 763–65 (2013) (using multilevel regression with poststratification); Julianna Pacheco, Using National Surveys to Measure Dynamic U.S. State Public Opinion: A Guideline for Scholars and an Application, 11 State Polit & Pol Q 415, 419–20 (2011) (using multilevel regression, imputation, poststratification, and simple moving average); Jeffrey R. Lax and Justin H. Phillips, How Should We Estimate Public Opinion in the States?, 53 Am J Polit Sci 107, 109–10 (2009) (using disaggregation and multilevel simulation modeling); Andrew Gelman and Jennifer Hill, Data Analysis Using Regression and Multilevel/Hierarchical Models 301–10 (Cambridge 2007) (discussing the use of multilevel logistic regressions to estimate state-level opinions from national polls); David K. Park, Andrew Gelman, and Joseph Bafumi, Bayesian Multilevel Estimation with Poststratification: State-Level Estimates from National Polls, 12 Polit Analysis 375, 376–77 (2004) (using multilevel logistic regression).
• 371See note 370.
• 372See Michael J. Barber, et al, Online Polls and Registration-Based Sampling: A New Method for Pre-election Polling, 22 Polit Analysis 321, 322 (2014).
• 373For further explanation of the points made in this paragraph, see Appendix A.
• 374Of course, what counts as an “opportunity district” will depend on one’s theory of dilution. Under the proportional representation theory, a district might be deemed an opportunity district only if the median (eligible?) voter in the district is close to the median voter of the minority community. Under the voter-discrimination theory, a district might be deemed an opportunity district if the median voter does not prefer white candidates to otherwise-similar minority candidates. Under the coalitional-breakdown theory, the district might be deemed an opportunity district if the median voter in the district prefers the platform of the minority-preferred umbrella coalition to the platform of the opposing coalition.
• 375Regarding these different approaches, see Reva B. Siegel, From Colorblindness to Antibalkanization: An Emerging Ground of Decision in Race Equality Cases, 120 Yale L J 1278, 1286–1303 (2011).
• 376Thanks to Professors Heather Gerken and David Schleicher for suggesting this point.
• 377For a general overview of issues in the estimation of racial bloc voting, see generally D. James Greiner, Ecological Inference in Voting Rights Act Disputes: Where Are We Now, and Where Do We Want to Be?, 47 Jurimetrics 115 (2007).
• 378See Christopher S. Elmendorf, Advisory Rulemaking and the Future of the Voting Rights Act, 14 Election L J 260, 268–75 (2015).
• 379See Bertrall L. Ross II, Denying Deference: Civil Rights and Judicial Resistance to Administrative Constitutionalism, 2014 U Chi Legal F 223, 228–29 (explaining that whether the judiciary defers to administrative agencies in the civil rights context depends on “judicial resistance to ‘administrative constitutionalism’”).
• 380The constraint would be illusory insofar as district courts remained free to reverse engineer whether the cutoff is met via discretionary decisions about which elections to include in or exclude from the polarization analysis. See Part II.B.2.c (explaining the discretion that courts currently exercise in this respect). It would be arbitrary if the appellate courts established bright-line rules about which elections to include and required equal weighting of those elections. See Part III.B (explaining the contingent relationship between votes and political preferences).
• 381See Shelby County, Alabama v Holder, 133 S Ct 2612, 2631 (2013).