This appeal involves a claim that the City of Martin, South Dakota, and several of its officials violated Section 2 of the Voting Rights Act of 1965, 42 U.S.C. § 1973(b). The plaintiffs contend that the defendants adopted and maintained an ordinance that impaired the ability of Native American Indians to participate in the political process and to elect representatives of their choice in city elections. Sitting en banc, we conclude that the district court properly dismissed the action in its order of March 22, 2005, which was reversed by a panel of this court. We therefore vacate the court’s later judgment of February 9, 2007, and remand with directions to dismiss the action.
I.
Pearl Cottier and Rebecca Three Stars, members of the Oglala Sioux Tribe and residents of Martin, brought suit against the City, several members of the city council, and the City’s former finance director, alleging violations of the Voting Rights Act and the Constitution. The plaintiffs alleged that the City’s Ordinance 122, which established boundaries for three voting wards within the City, diluted the votes of Indians in each ward, and thereby violated Section 2. They also alleged that the City enacted and maintained Ordinance 122 with a racially discriminatory purpose, in violation of Section 2 and the Fourteenth and Fifteenth Amendments.
After an eleven-day bench trial, the district court rejected the plaintiffs’ claims and dismissed the action. Cottier v. City of Martin, No. 02-5021, slip op. (D.S.D. Mar. 22, 2005) (hereafter “March 2005 Order ”). With respect to the Section 2 vote dilution claim, the court found that although the plaintiffs satisfied two of the three preconditions for liability that were established in Thornburg v. Gingles, 478 U.S. 30, 106 S.Ct. 2752, 92 L.Ed.2d 25 (1986), they failed to show the third precondition, namely, that the “white majority” in the City voted “sufficiently as a bloc to enable it ... usually to defeat the [Indian] preferred candidate.” Id. at 51, 106 S.Ct. 2752. The court also found no evidence of discriminatory intent in the passage of Ordinance 122, and dismissed the plaintiffs’ alternative Section 2 claim and the constitutional claims on that basis.
On appeal, a divided panel of this court reversed on the vote dilution claim. Cottier v. City of Martin, 445 F.3d 1113 (8th Cir.2006) (Cottier I). The court affirmed the district court’s findings regarding the first two Gingles preconditions, but concluded that the district court clearly erred in finding that the third precondition was not satisfied. The court remanded to the district court with instructions to determine whether, in view of this court’s ruling that the plaintiffs had met all three Gingles preconditions, the plaintiffs were entitled to relief under the totality of the circumstances. If so, the district court was directed to devise and implement an appropriate remedy. The City’s petition for rehearing en banc was denied, with five judges voting to grant it.
On remand, having been directed to accept that the plaintiffs established all three Gingles preconditions for a Section 2 vote dilution claim, the district court found based on the totality of the circumstances that Ordinance 122 violated Section 2. Cottier v. City of Martin, 466 F.Supp.2d 1175 (D.S.D.2006). The City declined to propose a remedy, asserting that there was no possible remedy for the violation found by the court. The district court considered three remedies proposed by the plaintiffs, and adopted the plaintiffs’ Plan C. Cottier v. City of Martin, 475 F.Supp.2d 932 (D.S.D.2007). Plan C did not divide the City into aldermanic wards, but rather *556adopted an at-large voting scheme using cumulative voting. Although the district court concluded in its March 2005 order that it lacked authority to order such a remedy, because it was not authorized by South Dakota law, see March 2005 Order at 21 n. 4 (citing Cane v. Worcester County, 59 F.3d 165, 1995 WL 371008 (4th Cir.1995) (unpublished), and Cane v. Worcester County, 35 F.3d 921 (4th Cir.1994)), the court on remand determined that Plan C was permissible. The court ruled that it was “bound to follow” dicta from this court’s opinion in Cottier I, 445 F.3d at 1123 n. 7, which stated that “[i]f, at the remedy stage, a redistricting of Martin’s wards appears unworkable, it appears that [Plan C] would be a viable option.” See 475 F.Supp.2d at 937.
The City appealed both the finding of a Section 2 violation and the remedy, and a divided panel of this court affirmed. Cottier v. City of Martin, 551 F.3d 733 (8th Cir.2008) (Cottier II). The court then granted rehearing en banc and vacated the panel opinion in Cottier II. The en banc court notified the parties that the court may wish to consider issues decided in Cottier I, as well as those briefed in Cottier II.
II.
As the case is before the en banc court for the first time, we must first consider the scope of our review. The present appeal arises from the district court’s rulings on remand from Cottier I, but this does not mean that we are constrained as a matter of law to accept the panel decision in Cottier I. The en banc court does not lightly review a prior panel decision in the same case, but we have the power to do so.
When sitting en banc, the court has authority to overrule a prior panel opinion, whether in the same case or in a different case. The en banc court has not considered the questions decided in Cottier I, and the law of the case does not preclude our consideration of those issues at this stage. The law of the case doctrine “expresses the practice of courts generally to refuse to reopen what has been decided,” but it is “not a limit to their power.” Messinger v. Anderson, 225 U.S. 436, 444, 32 S.Ct. 739, 56 L.Ed. 1152 (1912); see Christianson v. Colt Indus. Operating Corp., 486 U.S. 800, 817, 108 S.Ct. 2166, 100 L.Ed.2d 811 (1988). The doctrine, moreover, holds less sway with respect to an en banc court that is considering issues previously decided by a three-judge panel. That the court previously denied a petition for rehearing en banc is not controlling, because the decision to deny rehearing en banc is a pure exercise of discretion. It is not a ruling on the merits.
The parties have no justifiable expectation that a denial of rehearing en banc at an interlocutory stage resolves issues for all time. A remand order is not final until the Supreme Court denies certiorari at the end of the case. See Christianson, 486 U.S. at 817, 108 S.Ct. 2166 (“A petition for writ of certiorari can expose the entire case to review.”); Hughes Tool Co. v. Trans World Airlines, Inc., 409 U.S. 363, 365 n. 1, 93 S.Ct. 647, 34 L.Ed.2d 577 (1973) (holding that a prior dismissal of a writ of certiorari at an interlocutory stage did not establish law of the case or amount to res judicata on the points raised); E. Gressman et al., Supreme Court Practice 82-83 (9th ed. 2007) (“The Court can reach back and correct errors in the interlocutory proceedings below, even though no attempt was made at the time to secure review of the interlocutory decree or even though such an attempt was made without success.”). The parties likewise must recognize that when the court of appeals declines in its discretion to rehear *557a case en banc after a panel orders a remand, the court retains authority to rehear the matter en banc at a subsequent stage of the proceedings.
To the extent that a footnote in Robertson Oil Co. v. Phillips Petroleum Co., 14 F.3d 373, 376 n. 5 (8th Cir.1993) (en banc), purports to restrict the authority of an en banc court to consider prior panel decisions, we overrule it. Robertson cited law of the case principles that constrain a subsequent three-judge panel, see Liberty Mut. Ins. Co. v. Elgin Warehouse & Equip., 4 F.3d 567, 573 (8th Cir.1993) (order denying rehearing en banc), and seemed to apply them to an en banc court. We reject this view, and align ourselves instead with the uniform position of the circuits that an en banc court may overrule an erroneous panel opinion filed at an earlier stage of the same case. See Irving v. United States, 162 F.3d 154, 161 & n. 7 (1st Cir.1998) (en banc); Watkins v. U.S. Army, 875 F.2d 699, 704-05 n. 8 (9th Cir. 1989) (en banc); Shimman v. Int’l Union of Operating Eng’rs, Local 18, 744 F.2d 1226, 1229 n. 3 (6th Cir.1984) (en banc); Van Gemert v. Boeing Co., 590 F.2d 433, 437 n. 9 (2d Cir.1978) (en banc); In re Cent. R.R. Co. of N.J., 485 F.2d 208, 210-11 (3d Cir.1973) (en banc); see also 18B Charles Alan Wright, Arthur R. Miller & Edward H. Cooper, Federal Practice and Procedure § 4478.2 (2d ed.2002).
We conclude that it is appropriate in this case for the en banc court to consider the panel decision in Cottier I. For the reasons set forth below, we hold that Cottier I erred in reversing the district court’s dismissal of the action and remanding for further proceedings on liability and remedy. A federal court order directing a city to redraw its election ward boundaries, or to alter fundamentally its voting system, raises significant issues of federalism. See Voinovich v. Quitter, 507 U.S. 146, 156-57, 113 S.Ct. 1149, 122 L.Ed.2d 500 (1993). We think it exceptionally important for a federal court to ensure that there is a proven violation of Section 2 before ordering a city in South Dakota to undertake significant changes in its electoral process. Therefore, we will not overlook the error in Cottier I and proceed to contemplate an order directing the City of Martin to implement a remedy for a non-existent violation of the Voting Rights Act.2
III.
A.
In its March 2005 order, the district court considered the three preconditions for a Section 2 vote dilution claim, as described by the Supreme Court in Gingles:
*558First, the minority group must be able to demonstrate that it is sufficiently large and geographically compact to constitute a majority in a single-member district.... Second, the minority group must be able to show that it is politically cohesive.... Third, the minority must be able to demonstrate 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.
Gingles, 478 U.S. at 50-51, 106 S.Ct. 2752 (internal citation omitted).
On the first precondition, the district court found that two illustrative redistricting plans introduced by the plaintiffs showed that the Native American group in the city was sufficiently large and geographically compact to constitute a majority in a single-member district. The court rejected the City’s contentions that the majority was not large enough and that it was too fragile. The court also held that the proposed plans were not “so irregular on their face that they appear to be solely an effort to segregate races for the purposes of voting.” Rather, the court credited the testimony of the plaintiffs’ expert that he applied “traditional districting principles” to create these illustrative plans, and held that some consideration of race in fashioning the plans did not make them impermissible remedies for a Section 2 violation.
With respect to the second precondition, the court considered statistical analysis presented by experts for both parties concerning elections within the city and outside the city. The court found that three statistical methods employed by one or both experts were reliable, and that the analyses demonstrated political cohesiveness among Indians in Martin. Considering the statistical evidence together with historical evidence and contemporary lay testimony, the court found that the plaintiffs satisfied the second Gingles precondition.
Considering the third precondition, however, the district court found that the plaintiffs failed to meet their burden to show 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.” Gingles, 478 U.S. at 51, 106 S.Ct. 2752 (internal citation omitted). The court first evaluated an exit poll offered by the plaintiffs to show bloc voting in two city council races, but gave no weight to the poll due to shortcomings in its methodology.
The court then considered statistical analysis of election results from jurisdictions outside the city, which the court described as “exogenous” races. The court divided these exogenous races into several categories and made findings with respect to each. The court found that in countywide elections between candidates of different races, countywide elections between white candidates, and state and federal elections between white candidates, white voters did not vote sufficiently as a bloc usually to defeat the Indian-preferred candidate. In one category, statewide elections between candidates of different races, the court found that white voters did vote sufficiently as a bloc usually to defeat the Indian-preferred candidate. The court reasoned that interracial contests were entitled to more weight than contests among candidates of the same race, and that countywide elections should receive greater weight than state and federal elections. Finding that only one category of exogenous elections showed bloc voting by a white majority to defeat Indian-preferred candidates, the court conclud*559ed that the overall statistical evidence did not demonstrate the bloc voting required by Gingles.
The court also evaluated the testimony of lay witnesses who said that they could identify Indian-preferred candidates in city council elections, and that those candidates consistently lost. The plaintiffs urged that white bloc voting was the only possible explanation for the defeat of Indian-preferred candidates, but the district court found that the lay testimony did not eliminate other reasons for candidate losses. The court ultimately found that “[i]n light of the overwhelming statistical evidence, this lay testimony is not sufficient to meet plaintiffs’ burden of demonstrating the usual defeat of the Indian-preferred candidate.”
Having determined that the plaintiffs failed to satisfy the third Gingles precondition, the court entered judgment for the City on the plaintiffs’ claim of vote dilution in violation of Section 2. The plaintiffs appealed, and this court filed its opinion in Cottier I. We now consider the issues raised in that appeal.
B.
Vote dilution claims are “peculiarly dependent upon the facts of each case,” requiring “an intensely local appraisal of the design and impact of the contested electoral mechanisms.” Gingles, 478 U.S. at 79, 106 S.Ct. 2752 (internal quotations omitted). To “preserve[] the benefit of the trial court’s particular familiarity with the indigenous political reality,” id., we apply a “clear error” standard of review both to the predicate factual determinations and to the ultimate finding regarding vote dilution. Id.; Abrams v. Johnson, 521 U.S. 74, 91, 93, 117 S.Ct. 1925, 138 L.Ed.2d 285 (1997). It is the plaintiffs’ burden to demonstrate the existence of vote dilution. Voinovich v. Quilter, 507 U.S. 146, 155-56, 113 S.Ct. 1149, 122 L.Ed.2d 500 (1993).
We conclude that the district court did not commit clear error in finding that the plaintiffs failed to meet the third Gingles precondition, ie., that the white majority in Martin voted sufficiently as a bloc to enable it usually to defeat the minority’s preferred candidate in city council elections. We therefore overrule the panel decision in Cottier I.
First, the district court did not clearly err in giving no weight to the 2003 exit poll offered by the plaintiffs to show bloc voting by a white majority in city council elections. The court gave specific and cogent reasons for declining to credit the results of this poll. After observing that exit polls generally are “prone to high nonresponse rates which can seriously bias estimates and distort inferences,” March 2005 Order at 41, and that “exit poll respondents may lie,” id., the court found shortcomings in the specific exit poll conducted in Martin.
The court explained that the plaintiffs’ own expert admitted that the exit poll sample “under-represented non-Indians, over-represented Indians, and slightly over-represented females,” so that it was “not a representative sample of voters as a whole.” Id. The court found that the “high nonresponse rates of non-Indians seriously distorted inferences that could be drawn from the exit poll.” Id. at 42. The court also reasoned that “[a] truly representative poll of the votes actually cast should logically demonstrate some consistency between the responses to the poll and the actual returns,” id. at 41, but determined that the exit poll offered by the plaintiffs failed to demonstrate such consistency. Id. at 41-42. Although it was “troubled by the fact that some of the poll takers were related to one of the *560plaintiffs,” id. at 42, the court found it unnecessary to determine whether this fact alone was a sufficient basis to disregard the data.
It was not clear error for the court to view the poll as unreliable evidence of voting behavior by residents of Martin. There were reasonable grounds for the district court to believe that the poll results understated the number of Indians who favored the non-Indian-preferred candidate, understated the number of white voters who favored the Indian-preferred candidate, and failed to reflect truthful answers of those who responded.
Second, the plaintiffs’ challenge to the district court’s statistical analysis does not demonstrate clear error. The Cottier I panel concluded that the district court improperly relied on statewide and national election results in determining that the plaintiffs failed to meet their burden, but we disagree. Although the district court did use data from state and national elections, the district court did not rely on the state and national outcomes in tabulating the wins and losses of Indian-preferred candidates. Rather, the district court relied on the precinct-level results for those elections, which were reported by the plaintiffs’ expert and cited by the plaintiffs as the election results for the City of Martin.
The city election results derived from exogenous elections do not demonstrate that the district court’s overall finding as to the third Gingles factor was clearly erroneous. The number of interracial elections presented to the district court was very small. There were only four head-to-head interracial countywide races; one race was non-polarized, and the non-Indian-preferred candidate won the other three in the city. Appellants’ App. 555-56. There were only three statewide interracial head-to-head races, and the non-Indian-preferred candidate won those in the city. Appellants’ App. 555-56. But there were twenty-five state and federal races with white-only candidates, and the Indian-preferred candidate won fifteen of those contests in the city. Appellants’ App. 573-577. There were three white-only countywide races; one contest was non-polarized, and the Indian-preferred candidate was victorious in one of the other two in the city. Appellants’ App. 574-76. These results taken as a whole show almost equal numbers of victories for Indian-preferred candidates and non-Indian-preferred candidates. They do not compel a finding that a white majority in Martin votes sufficiently as a bloc usually to defeat the Indian-preferred candidate. Cf. Johnson v. Hamrick, 296 F.3d 1065, 1078 (11th Cir.2002) (“Although we have on various occasions held that district courts deciding [minority] vote dilution claims may give more weight to elections involving [minority] candidates than those involving all white contestants, there is no requirement that a district court must do so.”) (citations omitted).3
*561The district court did, for a subset of elections, rely on county election results, rather than city results, but this portion of the analysis does not establish clear error. The results in six of these contests, involving state and federal elections with multiple white candidates, actually favored the plaintiffs, because the Indian-preferred candidates prevailed in only two of the six. Appellants’ App. 559-60, 562, 564. Nine other contests in this subset were multicandidate races in Bennett County that involved both Indian and non-Indian candidates. The plaintiffs’ expert proffered these results as part of his analysis of the second Gingles precondition, but concluded that it was not possible to include these multi-candidate elections in his statistical analysis of the third precondition. Appellants’ App. 531-32. The City’s expert replied that the plaintiffs’ expert was “dead wrong” that these elections could not be incorporated into the analysis of the third precondition, Appellants’ App. 796, but the plaintiffs still submitted no city-level data concerning these elections. The district court evidently agreed with the City’s expert that the multi-candidate elections should be considered, and then evaluated the only data available in the record about those elections — ie., county-level results showing that the Indian-preferred candidates won five of the nine contests. March 2005 Order at 55-56; Appellants’ App. 547, 549-50. The court’s approach was more favorable to the plaintiffs than if the court had simply counted all nine elections against the plaintiffs based on a failure of proof.4
Overall, even if we accord little or no weight to the county results cited by the district court, the statistical evidence on the third Gingles precondition is mixed.5 The district court found insufficient proof, based on exogenous elections, that white voters typically vote as a bloc to defeat Indian-preferred candidates in the City. The data in the record fall short of demon-*562stating that this conclusion is clearly erroneous. See Gingles, 478 U.S. at 101, 106 S.Ct. 2752 (O’Connor, J., concurring in judgment) (concluding that even where the district court clearly erred by aggregating certain data and relying on it to find racial bloc voting, the district court’s ultimate conclusion was not clearly erroneous in light of other evidence that the court also considered).
Third, we disagree with the Cottier I panel that the district court clearly erred when it declined to consider the results of aldermanic elections from 2002-2004 as evidence of racial bloc voting. Other than the 2003 exit poll that the district court permissibly found unreliable, there was no statistical evidence regarding these elections. The only other evidence about these contests was the testimony of lay witnesses who expressed an opinion about which candidates were preferred by Indian voters, and whether those candidates won or lost. This testimony was disputed. Some witnesses presented by the plaintiffs did not even live in Martin. The defendants introduced testimony from Indian voters who did reside in Martin, and this evidence tended to show that Indians, in fact, have varied opinions on issues of the day and on preferred candidates for elective office. The district court considered the lay testimony, but found that it did not show that Indian-preferred candidates lost because of white bloc voting. The court concluded that in view of the statistical evidence, the testimony was insufficient to meet the plaintiffs’ burden to satisfy the third Gingles precondition. This is a factual finding that addresses the relative persuasiveness of disputed lay testimony and statistical evidence unfavorable to the plaintiffs. There is no clear error in the district court’s weighing of the evidence.
The record evidence in support of the plaintiffs’ case on the third Gingles precondition is not so strong as to generate a “definite and firm conviction” that the district court mistakenly dismissed the Section 2 vote dilution claim. See Anderson v. Bessemer City, 470 U.S. 564, 573, 105 S.Ct. 1504, 84 L.Ed.2d 518 (1985). For the reasons set forth by the district court, we also agree that the evidence is not sufficient to support a finding that Ordinance 122 was adopted and maintained for a discriminatory purpose. We therefore conclude that the district court’s judgment of March 22, 2005, should have been affirmed. Because the Section 2 vote dilution claim was properly dismissed based on the third Gingles precondition, we need not consider the City’s contentions regarding the other Gingles preconditions, the district court’s analysis of the totality of the circumstances on remand, or the permissibility of any remedies proposed by the plaintiffs. The panel opinion in Cottier I is set aside in its entirety, and it should not be treated as binding circuit precedent.
For the foregoing reasons, the district court’s judgment of February 9, 2007, is vacated, and the case is remanded with directions to dismiss.
SMITH, Circuit Judge, with whom MURPHY, BYE, and MELLOY, Circuit Judges, join, dissenting.
I respectfully dissent because I conclude that (1) this court should not, under the present circumstances, reconsider Cottier I and (2) we should grant the plaintiffs’ motion requesting that this court vacate the district court’s remedial order and remand to the district court for reconsideration of the appropriate remedy in light of the Supreme Court’s recent decision in Bartlett v. Strickland, — U.S. -, 129 S.Ct. 1231, 173 L.Ed.2d 173 (2009).
*563A. Reconsideration of Cottier I
In Robertson Oil, this court, sitting en banc, concluded that the court “should not” “accept a party’s motion for rehearing en banc of a decision after a second remand to open up the entire litigation.” 14 F.3d at 376 n. 5 (emphasis added). Specifically, the court observed that
[t]he dissent’s willing acceptance of this opportunity reveals a different view of the principles of the law of the case which a majority of the judges of this court so firmly endorsed in our order denying rehearing en banc in Liberty Mutual Insurance Co. v. Elgin Warehouse and Equipment, 4 F.3d 567, 573 (8th Cir.1993). Indeed, the dissent would take this rehearing of Robertson III to vacate Robertson I, in which we denied rehearing, and vacate Robertson II, in which we denied rehearing, and remand the case for a new trial.
Id.
The majority is correct that the view expressed in Robertson Oil is contrary to the prevailing view among our sister circuits that “[a] ruling by a panel ... does not ... establish the law of the case if a later appeal is heard by the court en banc.” 18B Charles Alan Wright, Arthur R. Miller & Edward H. Cooper, Federal Practice and Procedure § 4478.2 (2d ed.2002). The prevailing view maintains that
[a] court en banc surely is free to limit the matters it considers en banc, and may exercise its discretion to refuse reconsideration of some issues resolved by a panel on an earlier appeal. There is no reason why en banc reconsideration of all issues must follow the determination that one or more issues in a case warrant the grudgingly rationed resource of en banc consideration. All ordinary law-of-the-case concerns supplement this particular concern. Nonetheless, the court en banc has a special authority that supports a special freedom to redetermine important issues without feeling bound in the same way as successive panels.
Should successive appeals in the same case come to be heard by the court en banc, there is force in the view that ordinary law-of-the-case principles should apply. The en banc court, however, has a special authority and responsibility for the law of the circuit, and may properly assert an extra measure of freedom to reconsider important issues.
Id. (emphasis added) (footnote omitted).
But even if this court, sitting en banc, possesses the power to “overrule any panel decision that a majority of the active judges believes was wrongly decided,” Van Gemert v. Boeing Co., 590 F.2d 433, 437 n. 9 (2d. Cir.1978) (en banc), including in cases where prior en banc review of the panel’s decision was previously denied, the relevant question is under what circumstances the court should exercise this discretionary power.
I respectfully suggest that we should not exercise such power if “a party would be seriously prejudiced as a result,” id., or if reconsideration of a prior panel’s opinion would undermine “stability in the law — a sort of permanence and sureness in decision apart from the make-up or composition of the particular tribunal so far as the person of the Judges is concerned,” Lincoln Nat’l Life Ins. Co. v. Roosth, 306 F.2d 110, 113 (5th Cir.1962). As the Fifth Circuit explained:
In more tranquil days and times, an appeal from a second trial would be heard by the same Court as the first appeal. Now, that is highly unlikely, and where it occurs, it is — at least in this Court — due entirely to the laws of chance. That puts a premium on multi*564pie appeals. That is so because, without implying any improper purpose to litigants or their counsel, or acknowledging anything more than, as human beings, Judges will unavoidably have differences in emphasis, approach, or views on close questions in given areas, if the practice is followed for each succeeding panel to arrive at its own decisions, the losing party on the first appeal will naturally strive to bring it back a second, or a third, or a fourth time until all are exhausted. This possibility involves something other than simply more grist for our mill and as to which we should be indifferent.
* * *
We think that in a multi-Judge Court it is most essential that it acquire an institutional stability by which the immediate litigants of any given case, and equally important, the bar who must advise clients or litigants in situations yet to come, will know that in the absence of most compelling circumstances, the decision on identical questions, once made, will not be re-examined and re-decided merely because of a change in the composition of the Court or of the new panel hearing the case.
Id. at 114 (emphasis added).
In the present case, I conclude that
nothing about this case warrants our exercising the undoubted power to overrule the prior decision reached by the Court on the first appeal. On the contrary, any effort to re-examine the merits and now declare a result — either the same or a different one — independent of the former decision leads to consequences much more serious to the permanent, objective, administration of justice under law than any supposed individualized injustice to one or all of the litigants.
Id. First, neither party asked this court to reconsider the prior panel decision in Cot-tier I; instead, the en banc court sua sponte raised the question.
Second, no factual circumstances have changed since Cottier I was decided to justify reconsideration of the panel’s opinion; instead, a majority of the court— sitting three years later — -would merely apply the law to the same facts differently. We should not encourage parties to bring multiple appeals in the same litigation in the hope that the en banc court will overrule long-established prior panel opinions in that litigation. See Lincoln, 306 F.2d at 114.
Lastly, reconsideration of Cottier I unfairly prejudices plaintiffs. Plaintiffs commenced the instant litigation on April 3, 2002. Cottier, 466 F.Supp.2d at 1181. The district court’s decision rejecting plaintiffs’ claims was entered on March 22, 2005. See March 2005 Order. The appeal in Cottier I was orally argued on January 9, 2006, and the panel filed its opinion, as corrected, on May 8, 2006. This court denied rehearing en banc on June 28, 2006. Thus, at the time this case was orally argued to the en banc court on September 23, 2009, over three years elapsed since Cottier I was first submitted to this court. During this period of time, plaintiffs have functioned with the belief that they established the City’s § 2 liability under the VRA. No compelling circumstances, such as changed facts or law, justify reexamining and redeciding Cottier I. Indeed, applying Rule 35(a) of the Federal Rules of Appellate Procedure, this case, at its present posture, neither presents a necessity “to secure or maintain uniformity of the court’s decisions” nor “involves a question of exceptional importance.”6
*565B. Motion To Remand
Even though I would decline to reexamine Cottier I, I would not reach the merits of the City’s argument that the district court lacked the authority to impose remedial Plan C. Prior to oral argument, plaintiffs filed a motion requesting that this court vacate the district court’s remedial order and remand to the district court for reconsideration of the appropriate remedy in light of the Supreme Court’s recent decision in Bartlett, in which the Court declared that “the majority-minority rule relies on an objective, numerical test: Do minorities make up more than 50 percent of the voting-age population in the relevant geographic area?” 129 S.Ct. at 1245.
According to plaintiffs, the Supreme Court’s intervening decision in Bartlett alters the law on which the district court’s remedial order was premised. They cite to the district court’s belief that “[a]n effective majority, and thus an adequate remedy [to a § 2 claim], requires approximately 60 percent minority VAP.” Cottier, 475 F.Supp.2d at 938 (citing Jeffers v. Clinton, 756 F.Supp. 1195, 1198 (E.D.Ark. 1990)). Because the district court imposed a 60-pereent-minority-VAP requirement, plaintiffs contend that the district court erroneously rejected their proposed Plan A. Plan A provided for three duahmember districts, and its only majority-minority district had an Indian VAP of 54.55 percent. The district court concluded that “Plan A fail[ed] to provide an effective majority because it falls well short of the 60 percent VAP guideline.” Id. at 938. Plaintiffs also argue that, for the same reason, the district court improperly rejected Plan B, which proposed six single-member districts, with “Ward 1 ... containing] 53.51 percent Indian VAP” and “Ward 2 ... containing] 52.73 percent VAP.” Id. at 940.
In response, Martin maintains that Bartlett only discusses the liability stage of a § 2 case; it does not discuss an effective remedy or how a district court should determine remedies to a § 2 violation. According to Martin, Bartlett merely determined the minimum size minority group necessary to satisfy the first Gingles precondition in the liability phase. In support of its argument, Martin notes that even the dissenting justices in Bartlett emphasized that the Gingles preconditions do not state the ultimate standard under § 2. The dissent stated that “the threshold population sufficient to provide minority voters with an opportunity to elect their candidates of choice is elastic, and the proportions will likely shift in the future, as they have in the past,” citing that some courts use a requisite 65 percent majority-black population to constitute a safe district. Bartlett, 129 S.Ct. at 1254 (Souter, J., dissenting).
In this ease, the district court applied a per se rule that “[a]n effective majority, and thus an adequate remedy, requires approximately 60 percent minority VAP.” Cottier, 475 F.Supp.2d at 938. In light of Bartlett, I will consider the evolution of this per se rule and its post-Bartlett viability.
*5661. Pre-Bartlett Cases and Origins of the 60165-percent Rule
“In cases dealing with minority vote dilution, courts have confronted the question of what level of [minority] population is sufficient to provide the group with a ‘realistic opportunity to elect officials of their choice....'"Kimball Brace et al., Minority Voting Equality: The 65 Percent Rule in Theory and Practice, 10 Law & Pol’y 43, 45 (1988) (quoting Kirksey v. Bd. of Supervisors of Hinds County, 402 F.Supp. 658, 673, 676 (S.D.Miss.1975)). “This percentage has been called the ‘effective majority.”’ Id.
“[A] rule of thumb has evolved that sets a 65 percent minority population as the basis for an effective majority.” Id. The 65-percent rule was allegedly supported by the Justice Department and “given the imprimatur of the U.S. Supreme Court in United Jewish Organization of Williamsburgh v. Carey[, 430 U.S. 144, 97 S.Ct. 996, 51 L.Ed.2d 229] (1977).” Id. But “[n]either of these assertions is correct.” Id.
United Jewish Organization involved the Attorney General’s preclearance review of a redistricting plan under § 5 of the VRA. In that case, the Supreme Court stated that “[a]t a minimum and by definition, a ‘black majority district’ must be more than 50% black.” 430 U.S. at 162, 97 S.Ct. 996. The Court then determined that “it was reasonable for the Attorney General to conclude in this case that a substantial nonwhite population majority — in the vicinity of 65% — would be required to achieve a nonwhite majority of eligible voters.” Id. at 164, 97 S.Ct. 996. The Court’s conclusion was based on its determination that the size of the minority population in the minority districts must reflect “the need to take account of the substantial difference between the nonwhite percentage of the total population in a district and the nonwhite percentage of the voting-age population.” Id. at 163-64, 97 S.Ct. 996. Thus, the Court held that “the Justice Department had the authority ... to deny preclearance to a plan with insufficient Hispanic concentrations in certain districts in Brooklyn on the grounds that the plan failed to provide Hispanics an opportunity to elect candidates of then-choice.” Brace, supra, at 45.
Notably, the Court did not “suggest that the 65 percent rule is appropriate in all circumstances. Indeed, there is no discussion [in the case] of the empirical basis for the choice of the 65 percent figure.” Id.; see also Jack Quinn et al., Congressional Redistricting in the 1990s: The Impact of the 1982 Amendments to the Voting Rights Act, 1 Geo. Mason U. Civ. Rts. L.J. 207, 236 n. 120 (1990) (“The Court [in United Jewish Organizations ] did not otherwise explain or justify its acceptance of the 65% figure.”). Scholars have researched the origins of the 65-percent rule but have not found it in a holding of the Supreme Court.7
*567In 1986, the Supreme Court decided Gingles, the seminal § 2 VRA case. 478 U.S. at 30, 106 S.Ct. 2752. The Court determined that, to establish a § 2 violation, a plaintiff must prove that “a bloc voting majority must usually be able to defeat candidates supported by a politically cohesive, geographically insular minority group.” Id. at 49, 106 S.Ct. 2752. This test requires plaintiffs to make three separate showings. “First, the minority group must be able to demonstrate that it is sufficiently large and geographically compact to constitute a majority in a single-member district.” Id. at 50, 106 S.Ct. 2752. “Second, the minority group must be able to show that it is politically cohesive. 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.” Id. at 51, 106 S.Ct. 2752. “Third, the minority must be able to demonstrate 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, ... to defeat the minority’s preferred candidate.” Id.
Gingles did not explain “what the Court meant by the term ‘majority.’ It is unclear both as to what percentage Justice Brennan was referring to as being sufficient to constitute a majority for purposes of this test, and who could be included in calculating this figure.” Rick Strange, Application of Voting Rights Act to Communities Containing Two or More Minority Groups — When is the Whole Greater Than the Sum of the Parts?, 20 Tex. Tech. L.Rev. 95, 140 (1989). Gingles does not specify whether plaintiffs must show only a bare majority, whether they must show sufficient numbers so as to be able to effectively control a district, or who the court counts in calculating the majority. Id. The Court never discussed “whether the plaintiffs are required to constitute a majority of the proposed district’s total population, total voting age population, or total registered voters.” Id. Of course, the “ordinary meaning of majority is fifty percent plus one.” Id.
Post-Gingles, district courts within this circuit expressly applied a 60-percent or 65-percent rule. See, e.g., Smith v. Clinton, 687 F.Supp. 1361, 1363 (E.D.Ark.1988) (“A guideline of 65% of total population is frequently used, and is derived by supplementing a simple majority with an additional 5% to offset the fact that minority population tends to be younger than that *568of whites, 5% for the well-documented pattern of low voter registration, and 5% for low voter turnout among minorities. When voting-age population figures are used, a 60% nonwhite majority is appropriate.”); Jeffers, 756 F.Supp. at 1198 (“It was just this reasoning that led this Court, less than two years ago, to hold unanimously that something on the order of a 60% BVAP is required to remedy a vote-dilution violation of the Voting Rights Act”). And this court recognized that “ ‘[historically disadvantaged minorities require more than a simple majority in a voting district in order to have ... a practical opportunity to elect candidates of their choice.’ ” Whitfield v. Democratic Party of Ark., 890 F.2d 1423, 1428 (8th Cir.1989) (quoting Smith, 687 F.Supp. at 1362).
Nevertheless, “by 1990, the 65% rule was considered exceptional.” Richard H. Pildes, Is Voting-Rights Law Now at War with Itself? Social Science and Voting Rights in the 2000s, 80 N.C. L.Rev. 1517, 1527 (2002). “As one of the leading analyses concluded in the mid-1990s, districts with 55% [minority] populations were generally sufficient to enable [minority] voters to defeat racial bloc voting, while districts less than 45% [minority] almost never elected [minority] representatives.” Id. (citing David Lublin, The Paradox of Representation: Racial Gerrymandering and Minority Interests in Congress 46-47 (1997)). And, the Department of Justice “expressly disclaimed any reliance on a sixty-five percent standard.” Quinn, supra, at 239. On November 17, 1990, the chief of the Voting Section of the Civil Rights Division stated:
“Let me also take this opportunity to give you the Department of Justice’s position on the so-called ’65% Rule,’ i.e., whether a minority district must be at least 65% in order to satisfy Department of Justice requirements. We attach no particular significance to a 65% figure. The Department has frequently concluded, based on the facts presented in a particular submission, that districts containing a minority population significantly less than 65% (and even 50%) of the total may be entitled to Section 5 preclearance. We have also rejected plans where the minority percentage in a district exceeded 65%. Each Section 5 submission must be evaluated in light of the particular factual circumstances — not on the basis of some preordained population percentage.”
Id. (quoting Remarks of J. Gerald Hebert, Acting Chief, Voting Section, Civil Rights Division, Department of Justice at Conference on Fair Redistricting in Texas (Nov. 17, 1990)) (emphasis added by Quinn).
Despite the trend away from the 65-percent rule in the 1990s, in our most recent case regarding § 2 of the VRA, we continued to apply 65 percent as a “guideline.” Bone Shirt v. Hazeltine, 461 F.3d 1011, 1023 (8th Cir.2006). In Bone Shirt, we upheld the remedial plan adopted by the district court, finding that the plan “assures Native-American voters the opportunity to elect representatives of their choice and meaningful participation in the political process.” Id. We concluded that “the remedial plan affords Native-Americans more than a 65 percent majority in District 27 and a 74 percent majority in District 26A.” Id. In support of this holding, we cited Ketchum v. Byrne, 740 F.2d 1398, 1402 (7th Cir.1984), as acknowledging the 65-percent guideline and Neal v. Colebum, 689 F.Supp. 1426, 1438 (E.D.Va.1988), for the following proposition: “ ‘[T]he general 65% guideline for remedial districts is not a required minimum which the plaintiffs must meet before they can be awarded any relief under § 2.... Rather, the 65% standard is a flexible and practical guideline to consider *569in fashioning relief for a § 2 violation.’ ” Id. We also found that the district court “correctly considered other factors, including turnout rate and incumbency in formulating the plan.” Id. (holding that “all that is required [under § 2] is that the remedy afford Native-Americans a realistic opportunity to elect representatives of their choice”).
2. Bartlett v. Strickland
Did the Supreme Court’s decision in Bartlett modify what constitutes an “effective majority” for purposes of § 2? I conclude that it does. The case arose “in a somewhat unusual posture” in which the State of North Carolina invoked § 2 as a defense to a new district that it created. Bartlett, 129 S.Ct. at 1239. According to the State, “§ 2 required [it] to draw the district in question in a particular way,” despite a provision of the North Carolina Constitution that prohibited the state legislature from “dividing counties when drawing legislative districts for the State House and Senate.” Id. The Court granted certiorari to determine
whether [§ 2] can be invoked to require state officials to draw election-district lines to allow a racial minority to join with other voters to elect the minority’s candidate of choice, even where the racial minority is less than 50 percent of the voting-age population in the district to be drawn. To use election-law terminology: In a district that is not a majority-minority district, if a racial minority could elect its candidate of choice with support from crossover majority voters, can § 2 require the district to be drawn to accommodate this potential?
Id. at 1238. Ultimately, the Court held that such “crossover districts” do not satisfy the Gingles requirement that the minority population be large enough and yet sufficiently geographically compact to constitute a majority in a single-member district. Id. at 1243.
In reaching this holding, the Court began its analysis by noting that the case “turn[ed] on whether the first Gingles requirement can be satisfied when the minority group makes up less than 50 percent of the voting-age population in the potential election district.” Id. at 1241. The “dispositive question” was “[w]hat size minority group is sufficient to satisfy the first Gingles requirement?” Id. at 1242. The Court observed that
[a]t the outset the answer might not appear difficult to reach, for the Gingles Court said the minority group must “demonstrate that it is sufficiently large and geographically compact to constitute a majority in a single-member district.” 478 U.S., at 50, 106 S.Ct. 2752, 92 L.Ed.2d 25. This would seem to end the matter, as it indicates the minority group must demonstrate it can constitute “a majority.” But in Gingles and again in Growe [v. Emison, 507 U.S. 25, 113 S.Ct. 1075, 122 L.Ed.2d 388 (1993),] the Court reserved what it considered to be a separate question — whether, “when a plaintiff alleges that a voting practice or procedure impairs a minority’s ability to influence, rather than alter, election results, a showing of geographical compactness of a minority group not sufficiently large to constitute a majority will suffice.” Growe, supra, at 41, n. 5, 507 U.S. 25, 113 S.Ct. 1075, 122 L.Ed.2d 388; see also Gingles, supra, at 46-47, n. 12, 478 U.S. 30, 106 S.Ct. 2752, 92 L.Ed.2d 25. The Court has since applied the Gingles requirements in § 2 cases but has declined to decide the minimum size minority group necessary to satisfy the first requirement See Voinovich v. Quitter, 507 U.S. 146, 154, 113 S.Ct. 1149, 122 L.Ed.2d 500 (1993); [Johnson v.] De Grandy, [512 U.S. 997, 1009, 114 S.Ct. 2647, 129 L.Ed.2d 775 (1994)]; *570 League of United Latin American Citizens v. Perry, 548 U.S. 399, 443, 126 S.Ct. 2594, 165 L.Ed.2d 609 (2006) (opinion of KENNEDY, J.) (LULAC). We must consider the minimum-size question in this case.
Id. (emphasis added).
In rejecting the State’s argument that § 2 required the creation of crossover districts, the Court explained that permitting such claims
would require us to revise and reformulate the Gingles threshold inquiry that has been the baseline of our § 2 jurisprudence. Mandatory recognition of claims in which success for a minority depends upon crossover majority voters would create serious tension with the third Gingles requirement that the majority votes as a bloc to defeat minority-preferred candidates. It is difficult to see how the majority-bloc-voting requirement could be met in a district where, by definition, white voters join in sufficient numbers with minority voters to elect the minority’s preferred candidate.
Id. at 1244.
The Court “f[ou]nd support for the majority-minority requirement in the need for workable standards and sound judicial and legislative administration” because such a “rule draws clear lines for courts and legislatures alike.” Id. “Unlike any of the standards proposed to allow crossover-district claims, the majority-minority rule relies on an objective, numerical test: Do minorities make up more than 50 percent of the voting-age population in the relevant geographic area? ” Id. at 1245 (emphasis added). According to the Court, such a rule gives courts and officials charged with redistricting “straightforward guidance” as to what § 2 requires. Id.
Where an election district could be drawn in which minority voters form a majority but such a district is not drawn, or where a majority-minority district is cracked by assigning some voters elsewhere, then — assuming the other Gingles factors are also satisfied' — denial of the opportunity to elect a candidate of choice is a present and discernible wrong that is not subject to the high degree of speculation and prediction attendant upon the analysis of crossover claims. Not an arbitrary invention, the majority-minority rule has its foundation in principles of democratic governance. The special significance, in the democratic process, of a majority means it is a special wrong when a minority group has 50 percent or more of the voting population and could constitute a compact voting majority but, despite racially polarized bloc voting, that group is not put into a district.
Id. (emphasis added). Thus, the Court determined that it “remain[ed] the rule ... that a party asserting § 2 liability must show by a preponderance of the evidence that the minority population in the potential election district is greater than 50 percent.” Id. at 1246 (emphasis added). “The majority-minority rule ... is not at odds with § 2’s totality-of-the-circumstances test” because “the Gingles requirements are preconditions, consistent with the text and purpose of § 2, to help courts determine which claims could meet the totality-of-the-eircumstances standard for a § 2 violation.” Id. at 1247. According to the Court, “De Grandy confirmed ‘the error of treating the three Gingles conditions as exhausting the inquiry required by § 2.’ ” Id. (quoting De Grandy, 512 U.S. at 1013, 114 S.Ct. 2647).
3. Effect of Bartlett on Remedial Phase of § 2 Litigation
Admittedly, Bartlett concerned only the liability stage of a § 2 case, not the reme *571 dial stage. The question is whether Bartlett’s clarification of the minimum size minority group necessary to satisfy the first threshold Gingles requirement impacts what constitutes an “effective majority” in fashioning a remedy for a § 2 violation.
I find Bartlett instructive to the present case for two reasons. First, although the Court in Gingles and its progeny never definitively stated the minimum size minority group necessary for purposes of a § 2 vote-dilution claim, the Court in Bartlett did establish a numerical threshold and defined the term “majority.” The Court stated that, under the first Gingles factor, “the majority-minority rule” provides that a minority group must “make up more than 50 percent of the voting-age population in the relevant geographic area.” See Bartlett, 129 S.Ct. at 1245 (emphasis added). Thus, “a party asserting § 2 liability must show by a preponderance of the evidence that the minority population in the potential election district is greater than 50 percent.” Id. at 1246 (emphasis added).
Second, Bartlett’s explanation of the majority-minority rule with regard to liability directly affects the imposition of a § 2 remedy, as issues of liability and remedy are inextricably intertwined. See, e.g., Nipper v. Smith, 39 F.3d 1494, 1530-31 (11th Cir.1994) (en banc) (“The inquiries into remedy and liability ... cannot be separated: A district court must determine as part of the Gingles threshold inquiry whether it can fashion a permissible remedy in the particular context of the challenged system.”); accord Sanchez v. State of Colo., 97 F.3d 1303, 1311 (10th Cir.1996).
In addition to Bartlett, the per se 60-percent rule that the district court applied in the present case finds no reasoned basis in case law8 or statutory law, apart from the Court’s approval as “reasonable” the Attorney General’s decision that “a substantial nonwhite population majority — in the vicinity of 65% — would be required to achieve a nonwhite majority of eligible voters.” United States Jewish Organization, 430 U.S. at 164, 97 S.Ct. 996. “[T]he 65% benchmark is an artificial one because it does not necessarily reflect a minority group’s true potential to control a district.” Angelo N. Ancheta & Kathryn K. Imahara, Multi-Ethnic Voting Rights: Redefining Vote Dilution in Communities of Color, 27 U.S.F. L.Rev. 815, 867 (1993).
Therefore, in light of the 50-percent numerical threshold requirement in Bartlett, the interrelation between remedy and liability, and the lack of reasoned authority for imposing a 60-percent or 65-percent per se rule, I would remand to the district court for reconsideration of plaintiffs’ Plan A and Plan B. I would direct the district court to gather any additional statistical evidence, evaluate such evidence, and conduct a particularized inquiry to determine what percentage of minority voters is necessary to provide such voters with a reasonable opportunity to elect a representative of their choice. In conducting this inquiry, I would advise the district court to remain mindful that liability and remedy are inextricably intertwined. While 60 to 65 percent may, in some cases, constitute a sufficient minority population, it does not necessarily follow that such percentages also constitute the minimum sufficient percentage in every case. Under Bartlett, a minority need only make up “more than 50 percent of the voting-age population in the relevant geographic area” to satisfy the first Gingles factor. 129 S.Ct. at 1245 (emphasis added).
*572C. Conclusion
Accordingly, I would grant plaintiffs’ motion to vacate the district court’s remedial order and remand to the district court for reconsideration of a proper remedy in accordance with the foregoing instructions.