HEANEY, Circuit Judge.
Martin, South Dakota is a city of 1,078 people located adjacent to the Pine Ridge and Rosebud Indian Reservations. The city is divided into three, dual member wards, each of which elects its two aider-men every two years in staggered terms. Native-Americans make up nearly 45% of the total population and 36% of the voting-age population. Only twice since 1984 has an Indian-preferred candidate been elected alderman. In each case, their election was uncontested.
The American Civil Liberties Union brought an action on behalf of two Native Americans in district court, challenging the 2002, 2003, and 2004 elections. The complaint alleged that the city wards were configured in a manner that intentionally and effectively diluted the voting strength of Native-Americans and kept Indian-preferred aldermen candidates from being elected, contrary to the provisions of the Voting Rights Act of 1965 and the Fourteenth and Fifteenth Amendments to the United States Constitution. The district court denied relief, concluding that the white majority did not usually vote in a way to defeat the Indian-preferred candidate. We disagree and remand the matter to the district court to complete the analysis required by the United States Supreme Court pursuant to section 2 of the Voting Rights Act1 as construed by Thornberg v. Gingles, 478 U.S. 30, 49-50, 106 S.Ct. 2752, 92 L.Ed.2d 25 (1986). If the district court then finds in favor of the plaintiffs, it shall develop a plan under which Native-Americans will have a reasonable opportunity to elect an Indian-preferred candidate.
BACKGROUND
For more than a decade Martin has been the focus of racial tension between Native-Americans and whites. In the mid-1990’s, protests were held to end a racially offensive homecoming tradition that depicted Native-Americans in a demeaning, stereotypical fashion. Concurrently, the United States Justice Department sued and later entered into a consent decree with the local bank requiring an end, to “redlining” loan practices and poli[1116]*1116cies that adversely affected Native-Americans, and censuring the bank because it did not employ any Native-Americans. Most recently, resolution specialists from the Justice Department attempted to mediate an end to claims of racial discrimination by the local sheriff against Native-Americans.
With these conflicts as a background, Martin redrew the city’s wards because population shifts had rendered the existing boundaries obsolete. After the new wards were drawn and published as Ordinance 121, attorneys for a Native-American public interest group alleged the new boundaries violated the one-person, one-vote principle, see Reynolds v. Sims, 377 U.S. 533, 562-63, 84 S.Ct. 1362, 12 L.Ed.2d 506 (1964), and that the boundaries violated section 2 of the Voting Rights Act.
In March 2002, new districts were drawn to address the one-person, one-vote violation. After review by the South Dakota Attorney General’s office, the city council adopted the new ward boundaries and implemented them under Ordinance 122.2 After a failed attempt under South Dakota law to refer the ordinance to a voter referendum, see South Dakota Codified Laws § 9-20-6, plaintiffs initially brought suit alleging Ordinance 121 violated the Equal Protection Clause of the Fourteenth Amendment. The district court dismissed the complaint as moot because that ordinance had been repealed by Ordinance 122, but the court also granted plaintiffs motion to supplement its complaint to include the allegations currently pending before this court.
In March 2005, the district court entered a final judgment disposing of all of the parties’ claims. It found that although the plaintiffs met the first two conditions of Gingles, the plaintiffs failed to prove, by a preponderance of the evidence, the third Gingles precondition. As a result, the district court concluded the plaintiffs could not prevail in their vote dilution claim. Additionally, the court concluded that, since there was not sufficient evidence to prove a vote-dilution or “effects” claim, the plaintiffs also could not prove that the city of Martin adopted and maintained Ordinance 122 for a discriminatory purpose.
ANALYSIS
“The district court’s findings regarding the factual context ... are reviewed for clear error.” Harvell v. Blytheville Sch. Dist., 71 F.3d 1382, 1386 (8th Cir.1995) (en banc). Legal conclusions, “ ‘including those that may infect a so-called mixed finding of law and fact, or a finding of fact that is predicated on a misunderstanding of the governing rule of law,’ are subject to [de novo] review.” Id. (quoting Gingles, 478 U.S. at 79, 106 S.Ct. 2752).
“The essence of a § 2 claim is that a certain electoral law, practice, or structure interacts with social and historical conditions to cause an inequality in the opportunities enjoyed by black and white voters to elect their preferred candidates.” Gingles, 478 U.S. at 47, 106 S.Ct. 2752. Section 2 of the Voting Rights Act provides that a denial of the right to vote occurs when:
based on the totality of the circumstances, it is shown that the political processes leading to nomination or election in the State or political subdivision are not equally open to participation by members of a class of citizens protected by subsection (a) of this section in that its members have less opportunity than other members of the electorate to par[1117]*1117ticipate in the political process and to elect representatives of their choice.
42 U.S.C. § 1973(b).
The Supreme Court in Gingles established three preconditions to establishing a section 2 claim:
1) that the minority group is large enough and geographically compact enough that it would be a majority in a single-member district; 2) that the minority group is politically cohesive [in the sense that its members vote in a similar fashion]; and 3) that the white majority votes sufficiently as a bloc to enable it, in the absence of special circumstances, [to] usually ... defeat the minority’s preferred candidate.
Harvell, 71 F.3d at 1385 (citing Gingles, 478 U.S. at 50-51, 106 S.Ct. 2752). If the three Gingles preconditions are met, the court then considers the totality of the circumstances. Id. at 1390. Failure to establish all three preconditions defeats a section 2 claim. Clay v. Bd. of Educ., 90 F.3d 1357, 1362 (8th Cir.1996)
The district court found that the plaintiffs met the first two, but failed to meet the third, Gingles preconditions. We agree with the district court that the plaintiffs fulfilled the first two preconditions. As to the third precondition, we hold the plaintiffs proved by a preponderance of the evidence that the white majority votes as a bloc to usually defeat Indian-preferred candidates.
To establish the first Gingles precondition, a plaintiff must demonstrate a proper and workable remedy exists. Stabler v. County of Thurston, 129 F.3d 1015, 1025 (8th Cir.1997). The plaintiffs in this case offered three redistricting plans. The first redistricts Martin into three wards and creates at least one Native-American majority ward. The second redistricts Martin into six wards and creates at least two Native-American majority wards. Both would allow Native-Americans in Martin a more reasonable opportunity to place two representatives on the Martin city council. The third plan eliminates all ward boundaries and implements a system in which three members of the city council would be elected in each election. According to expert testimony, this would allow Native-Americans a reasonable opportunity to elect at least two representatives to the Martin city council. The district court found that the first two plans presented proper and workable remedies that could be implemented to alleviate the inequalities of the current system, but determined that it lacked the authority to authorize the at-large remedy proposed in the third plan.
Martin disagrees with the district court and argues that the first two plans presented by the plaintiffs are not viable or stable. We agree with the district court. The ultimate viability and effectiveness of a remedy is considered at the remedial stage of litigation and not during analysis of the Gingles preconditions. At the initial stage, the plaintiff must only show that “minority voters possess the potential to elect representatives in the absence of the challenged structure or practices.” Gingles, 478 U.S. at 50 n. 17, 106 S.Ct. 2752; see also Dickinson v. Indiana State Election Bd., 933 F.2d 497, 503 (7th Cir.1991); Houston v. Lafayette County, 56 F.3d 606, 611 (5th Cir.1995). Although Martin argues the plans are not viable or stable, the ultimate end of the first Gingles precondition is to prove that a solution is possible, and not necessarily to present the final solution to the problem. Gingles, 478 U.S. at 50 n. 17, 106 S.Ct. 2752.
The equal protection clause is violated if race is the predominant factor motivating the placement of a significant number of voters within or without a particular district. Stabler, 129 F.3d at 1025 (stating that bizarrely shaped districts [1118]*1118“considered in combination with racial and population densities of the proposed districts, support [a] finding that race was the predominant factor in drawing proposed districts to create a majority-minority single-member district”). As the district court noted, examples of bizarrely shaped districts that should raise concern include those that look like “a sacred Mayan bird, with its body running eastward ... [sjpindly legs reach south ... while the plumed head rises northward ... an open beak appears to be searching for worms.” (Appellants’ Add. at 19 (quoting Bush v. Vera, 517 U.S. 952, 974, 116 S.Ct. 1941, 135 L.Ed.2d 248 (1996) (additional citations omitted)).) “[Section] 2 compactness inquiry should take into account ‘traditional districting principles such as maintaining communities of interest and traditional boundaries.’ ” Abrams v. Johnson, 521 U.S. 74, 92, 117 S.Ct. 1925, 138 L.Ed.2d 285 (1997) (quoting Bush, 517 U.S. at 977, 116 S.Ct. 1941); Miller v. Johnson, 515 U.S. 900, 916, 115 S.Ct. 2475, 132 L.Ed.2d 762 (1995) (recognizing “respect for political subdivisions or communities defined by actual shared interests” as a traditional, race-neutral districting principle); Shaw v. Reno, 509 U.S. 630, 651-52, 113 S.Ct. 2816, 125 L.Ed.2d 511 (1993) (recognizing population equality as a “sound” districting principle).
In this case, we agree that the shapes of the proposed districts would not draw constitutional scrutiny because the districts are not primarily based on race. The record reflects that the proposed plans follow census blocks, marked streets, exhibited more than point contiguity, created wards of equal population, and recognized traditional neighborhoods. Moreover, Martin has maintained a ward system for at least forty years, proving its stability. Furthermore, according to census figures used at trial, it is highly unlikely that a Native-American majority district would fail to maintain its majority over time because the Native-American population in Martin is increasing rather than decreasing.3 For these reasons, we affirm the district court’s finding that the Native-American community in Martin is sufficiently large and geographically compact to constitute a majority in a single-member district.
To satisfy the second Gingles precondition, the plaintiff must demonstrate that the minority group 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.” Gingles, 478 U.S. at 51, 106 S.Ct. 2752. Evidence of political cohesiveness is shown by minority voting preferences, distinct from the majority, demonstrated in actual elections, and can be established with the same evidence plaintiffs must offer to establish racially polarized voting, because “political cohesiveness is implicit in racially polarized voting.” Sanchez v. Colorado, 97 F.3d 1303, 1312 (10th Cir.1996).
Proving political cohesiveness requires evaluating elections through statistical and non-statistical evidence. Cf. Growe v. Emison, 507 U.S. 25, 41, 113 S.Ct. 1075, 122 L.Ed.2d 388 (1993) (finding the district court erred in concluding political cohesiveness was proven where it was unsupported by statistical or anecdotal evidence). The district court relied on three proven approaches to evaluating elections: homogenous precinct analysis, bivariate ecological regression analysis, and ecological inference. See Rural W. Tenn. African-American Affairs Council v. Sundquist, 209 F.3d 835, 839 (6th Cir.2000) [1119]*1119(considering homogenous precinct analysis and bivariate ecological regression analysis); Teague v. Attala County, 92 F.3d 283, 290 (5th Cir.1996) (holding that the district court erred by disregarding “the established acceptance of regression analysis as a standard method for analyzing racially polarized voting”).
We agree with the district court that the record is clear that the statistical and non-statistical analysis proved political cohesion within the Native-American community. Statistically, experts from both sides found that the Native-American population on average voted for the same candidates more than 60% of the time. See African American Voting Rights Legal Def. Fund, Inc. v. Villa, 54 F.3d 1345, 1353 n. 11 (8th Cir.1995) (noting that “60% majority is merely a guideline, not an absolute threshold,” for finding political cohesion). Lay testimony demonstrates that Native-Americans in Martin and the surrounding area were politically cohesive. Examples of cohesion include political protests against the abuse of Native-American rights, the endorsement of a slate of Native-American political candidates, and the use of tribal governments to confront social issues such as education and housing.
Martin argues that if Native-Americans are politically cohesive, it is because of their political partisanship, and not because of racial identity. Although potentially relevant in the totality of the circumstances analysis, the reason for the cohesion is irrelevant in the threshold determination of whether the Gingles preconditions are met. Goosby v. Town Bd., 180 F.3d 476, 493 (2d Cir.1999) (interpreting Gingles as treating “causation as irrelevant in the inquiry into three Gingles preconditions”); Lewis v. Alamance County, 99 F.3d 600, 615 n. 12 (4th Cir.1996) (same). To imply that party affiliation should negate political cohesion would have the effect of denying minority voters an equal opportunity to elect representatives of their choice regardless of the reason. Goosby, 180 F.3d at 495-96. For these reasons, we affirm the district court’s decision that the Native-American community in Martin is a politically cohesive minority group.
A racial voting bloc “exists where there is a consistent relationship between the race of the voter and the way in which the voter votes, or to put it differently, where black voters and white voters vote differently.” Gingles, 478 U.S. at 53 n. 21, 106 S.Ct. 2752 (internal quotation marks and citations omitted). To be legally significant for the purposes of Gingles, the plaintiff must show that “whites vote sufficiently as a bloc usually to defeat the minority’s preferred candidates.” Id. at 56, 106 S.Ct. 2752; see also Sanchez, 97 F.3d at 1319. “The correct question is not whether white voters demonstrate an unbending or unalterable hostility to whoever may be the minority group’s representative of choice, but whether, as a practical matter, the usual result of the bloc voting that exists is the defeat of the minority-preferred candidate.” Jenkins v. Red Clay Consol. Sch. Dist. Bd. of Educ., 4 F.3d 1103, 1123 (3d Cir.1993). “[T]he presence of racially polarized voting will ordinarily be the keystone of a vote dilution case.” Buckanaga v. Sisseton Indep. Sch. Dist., 804 F.2d 469, 473 (8th Cir.1986).
To determine whether the white majority voted as a bloc to defeat the Indian-preferred candidate, three inquiries are required. First, who are the minority-preferred candidates?4 Old Person v. Cooney, 230 F.3d 1113, 1122 (9th Cir.2000). [1120]*1120Second, did the white majority vote “as a bloc to defeat the [minorityj-preferred candidate”? Id. And third, were there special circumstances, “such as the minority candidate running unopposed,” present when minority-preferred candidates won? Gingles, 478 U.S. at 51, 106 S.Ct. 2752.
In our view the district court erred in three respects when it determined the white majority usually did not defeat the minority-preferred candidate. First, it did not give sufficient weight to the exit poll of the 2003 elections. Second, it did not give any weight to the results of the 2002, 2003, and 2004 aldermanic elections. Finally, the district court relied exclusively on county, state, and national elections in determining that the evidence failed to prove that Indian-preferred candidates usually lost elections.
First, the district court rejected the exit poll results of the 2003 aldermanic races in Wards I and III. In Ward I, the Indian-preferred candidate received an adjusted 25% of the white vote and 100% of the Native-American vote. In Ward III, the Indian-preferred candidate received almost 37% of the white vote and almost 86% of the Native-American vote. The district court rejected these results because it believed the poll failed to ascertain a representative sample of the voters as a whole and because there was inconsistency between the poll results and the actual returns.
In our view, it was clear error to reject these statistics. Although the exit poll probably under-represents white voters and likely over-represents Native-American voters, the results show that despite a cohesive political effort by Native-American voters, their candidates for alderman were defeated. We find this, at the very least, probative. See Johnson v. De Grandy, 512 U.S. 997, 1011-12, 114 S.Ct. 2647, 129 L.Ed.2d 775 (1994); Jenkins, 4 F.3d at 1126 (holding that the third Gingles factor “may be satisfied with a variety of evidence, including lay testimony or statistical analysis of voting patterns”). Because the record reflects that Native-Americans and whites are, by and large, the only racial groups in Mai’tin, the only conceivable explanation for the results of the exit poll and the final election tallies is that the white majority voted as a bloc against Indian-preferred candidates. See Buckanaga, 804 F.2d at 473 (stating that “[t]he surest indication of race conscious politics is a pattern of racially polarized voting extending over time”); Collins v. City of Norfolk, 816 F.2d 932, 935 (4th Cir.1987) (holding that racially polarized voting patterns can establish “both cohesiveness of the minority group and the power of white bloc voting to defeat the minority’s candidates”).
Second, the district court ignored the results of the 2002, 2003, and 2004 alder-manic elections. The record clearly reflects that in 2002, the Indian-preferred candidates for alderman in each of the three wards lost. In 2003, the Indian-preferred candidates for alderman in Wards I and III lost. And in 2004, the Indian-preferred candidates for alderman in each of the wards lost.
The district court clearly erred when it ignored these election results. The plaintiffs presented eight aldermanic elections over the span of three years that established a sufficient pattern of defeat for Indian-preferred candidates in Martin’s aldermanic elections. See Gingles, 478 U.S. at 58-59, 80-82, 106 S.Ct. 2752 (finding data from three election years, involving minority candidates, sufficient to uphold district court’s vote-dilution ruling); Citizens for a Better Gretna v. City of Gretna, 834 F.2d 496, 502-03 (5th Cir.1987), cert. denied, 492 U.S. 905, 109 S.Ct. 3213, 106 L.Ed.2d 564 (1989) (stating that “Gingles ... suggests flexibility in the face of [1121]*1121sparse data”). Importantly, it is these elections-the Martin aldermanic elections-that are at the center of this litigation. Clark v. Calhoun County, 88 F.3d 1393, 1397 (5th Cir.1996) (holding that virtual absence of electoral success in relevant district has the greatest impact on the evaluation of vote-dilution claims). In short, the absence of Native-American aldermen, combined with evidence of racially polarized voting, provides striking proof of vote dilution in Martin.
Third, the district court found that the results of county, state, and national elections supported the view that Indian-preferred candidates did not usually lose elections in Martin. At the national level, the district court factored in the 1998 and 2002 races for United States Senate and House of Representatives as indications that Indian-preferred candidates were able to win elections. At the state level, victories by the Indian-preferred candidate in the 2002 elections for state senate district 26, state house district 26, state auditor, state secretary of state, and state treasurer were considered as part of the analysis of the third Gingles precondition. The district court also used county elections such as the 2002 races for county commissioner, sheriff, and school board as examples of Indian-preferred candidates who won elections. The court held these elections were evidence that the Indian-preferred candidates did not usually lose their elections in spite of the white majority bloc.
In our view, these elections provide very little evidence of whether Martin’s ward system allows Native-Americans to elect their preferred candidates. Although it can be appropriate to factor in exogenous elections,5 these elections are meant to supplement, not replace, endogenous elections. See Clay, 90 F.3d at 1362. The data gained from state and national elections did little to reveal whether there was racial polarization within the city of Martin’s ward system. For example, citing Tom Daschle’s 1998 victory in the United States Senate race may prove the relevance of the Native-American vote throughout the state, but Daschle’s victory fails to reveal whether Martin’s ward system allows Native-American residents to elect their preferred candidates to aider-man. Accord Westwego Citizens for Better Gov’t v. City of Westwego, 872 F.2d 1201, 1209 n. 11 (5th Cir.1989) (stating that “evidence derived from exogenous elections ... must be evaluated according to its probative value”). Because of their far-reaching scope, state and national elections offered little probative insight into whether Martin’s ward system violated section 2 of the Voting Rights Act. The same can be said for election results from Bennett County, where Martin is located. Indian-preferred candidates fare better in the county simply because Native-Americans make up 49.25 percent of the county’s voting-age population while whites make up 49.65 percent of the county’s voting age population. See id. at 1209-10 (focusing its analysis on the exogenous data from the precincts of the jurisdiction directly at issue).6
[1122]*1122Considering the entirety of evidence presented in this case, we hold that the plaintiffs proved by a preponderance of the evidence that the white majority usually defeated the Indian-preferred candidate in Martin aldermanic elections. First, the 2003 exit poll clearly showed racial polarization. See De Grandy, 512 U.S. at 1011, 114 S.Ct. 2647 (recognizing that the “ultimate conclusions about equality or inequality of opportunity were intended by Congress to be judgments resting on comprehensive, not limited, canvassing of relevant facts”); Harvell, 71 F.3d at 1386 (considering statistical analysis, exit polling, and lay testimony). Although Native-Americans predominantly voted for the Indian-preferred candidates, those candidates lost when the actual votes were counted. Because Native-Americans and whites make up more than 99% of Martin’s population, the only conclusion available is that whites voted as a bloc to defeat the Indian-preferred candidates. See Buckanaga, 804 F.2d at 473.
Secondly, whereas the district court did not give the aldermanic election results of 2002 through 2004 probative value, we consider these elections central to the analysis of whether there was a section 2 violation. The plaintiffs proved the third Gingles precondition in part because the aldermanic election results of 2002 through 2004 reveal the Indian-preferred candidates for alderman always lost these elections. See Gingles, 478 U.S. at 58-59, 80-82, 106 S.Ct. 2752; Jenkins, 4 F.3d 1103, 1123 (recognizing the most important aspect of the analysis is to determine whether the white voting bloc usually results in the defeat of the minority-preferred candidate). Finally, county, state, and national election results were not probative in determining whether Martin’s ward system denied Native-Americans a reasonable opportunity to elect their preferred candidates.
CONCLUSION
For each of the reasons noted herein, we reverse the decision of the district court and find that the plaintiffs met, by a preponderance of the evidence, all three Gingles preconditions. Thus, we remand the matter to the district court with instructions to initially determine whether, in view of the fact that plaintiffs have met all three Gingles preconditions, the plaintiffs are entitled to relief in light of the totality of the circumstances. The Supreme Court has listed the following factors as relevant in the totality of the circumstances analysis:
(1) the history of voting-related discrimination in the state or political subdivision; (2) the extent to which voting in the state or subdivision is racially polarized; (3) the extent to which the state or subdivision has used voting practices or procedures that tend to enhance opportunities for discrimination against the minority group; (4) whether minority candidates have been denied access to any candidate-slating process; (5) the extent to which minorities have borne the effects of past discrimination in relation to education, employment, and health; (6) whether local political campaigns have used overt or subtle racial appeals; (7) the extent to which minority group members have been elected to public office in the jurisdiction; (8) whether there is a significant lack of responsiveness on the part of elected officials to the particularized needs of members of the minority group; and (9) whether the policy underlying the use of voting qualifications is tenuous.
Harvell, 71 F.3d at 1385-86 (citing Gingles 478 U.S. at 36-37, 106 S.Ct. 2752). This court has previously focused the totality of the circumstances analysis on racial polarization and the ability to elect minority-[1123]*1123preferred candidates under the challenged scheme. Id. at 1390.
Plaintiffs do not need to prove a particular number of the above-listed factors or prove that a majority of them point one way or the other. Rather, “the final determination of whether the voting strength of minority voters is canceled out demands the court’s overall judgment, based on the totality of the circumstances and guided by those relevant factors in the particular case.” Whitfield v. Democratic Party, 890 F.2d 1423, 1432 (8th Cir.1989) (quoting S.Rep. No. 417, at 29 n. 118) (internal quotations and modifications omitted).
In the event the district court finds that under the totality of the circumstances, the plaintiffs are entitled to relief, the district court shall devise and implement a remedy that will give Native-Americans in Martin a reasonable opportunity to elect Indian-preferred candidates to alderman. In so doing, the defendant shall be given an opportunity to propose a remedy within a specified amount of time. See Cane v. Worcester County, 35 F.3d 921, 927 (4th Cir.1994). The court should then review the proposed order to determine whether it is “legally unacceptable.” Id. If the defendant fails to propose a legally acceptable remedy, the district court shall devise a plan that ensures that Indian-preferred candidates have a reasonable chance of prevailing in Martin municipal elections for alderman. Among its options, the district court has the discretion to implement any of the three plans presented by the plaintiffs.7