In my view, the court relies upon inadequate and unreliable statistical evidence to justify disregarding “consistent and sustained success by [African-American] candidates [that] is presumptively inconsistent with the existence of a § 2 violation.” Thornburg v. Gingles, 478 U.S. 30, 102, 106 S.Ct. 2752, 2793, 92 L.Ed.2d 25 (1986) (O’Connor, J. concurring). Therefore, I respectfully dissent.
A. Insufficient Evidence of Racially Polarized Voting
The district court found that plaintiffs failed to prove the third Gingles precondition, that the white majority in the Blythe-ville School District “votes sufficiently as a bloc to enable it — in the absence of special circumstances ... usually to defeat the minority’s preferred candidate.” 478 U.S. at 51, 106 S.Ct. at 2766-67. To qualify as “legally significant white bloc voting,” the Supreme Court explained in Gingles, the white vote must be sufficiently polarized that it will “normally ... defeat the combined strength of minority support plus white ‘crossover’ votes.” 478 U.S. at 56, 106 S.Ct. at 2769. In overturning the district court’s ultimate finding that the third precondition was not proved, this court relies primarily on a subsidiary finding that voting in Blytheville School Board elections has been “highly polarized.” Ante at 1390. That is a clearly erroneous finding, improperly made in the first instance by this court, not the district court.
At trial, plaintiffs relied on statistical analysis by their expert, James R. Lynch, a Senior Research Specialist in the Institute of Government at the University of Arkansas at Little Rock. As Appendix I reflects, Lynch submitted District-wide vote totals for every Blytheville School Board election from 1969 to 1992 that included an African-American candidate. He also presented vote totals by voting district for ten of those elections, in 1982, 1987, 1988, 1989, 1990, and 1991. Beyond that, all of Lynch’s data were in percentage form. He calculated the black voting age population percentage (“BVAP”) for each voting district, based on 1980 census data. He calculated the percent of the vote *1392captured by African-American candidates in each voting district in the ten selected races. He then employed a “correlation coefficient” to conclude that there was an “overwhelming relationship ... between BVAP and the percent of vote that the black candidates got.” See Appendix II. Lynch did not perform an extreme case analysis or a complete bivariate ecological regression analysis, methods the Supreme Court noted are “standard in the literature for the analysis of racially polarized voting.” Gingles, 478 U.S. at 53 n. 20, 106 S.Ct. at 2767 n. 20. Thus, the record here contains far less probative statistical data than was developed in other recent Voting Rights Act cases such as National Ass’n for the Advancement of Colored People, Inc. v. City of Niagara Falls, 65 F.3d 1002, 1005-06 & nn. 2-4 (2d Cir.1995), and Clay v. Board of Educ., 896 F.Supp. 929, 934-36 (E.D.Mo.1995) ,1
Lynch’s statistical analysis was more than inadequate, it was faulty. His most serious mistake was in equating racially cohesive voting and racially polarized voting, an equation contrary to Gingles. Racial cohesiveness and racial polarization are elements of different Gingles preconditions. Political cohesiveness is the second Gingles precondition — “a showing that a significant number of minority group members usually vote for the same candidates.” 478 U.S. at 56, 106 S.Ct. at 2769. I agree with the court that Lynch’s data tend to show that African-American voters have been politically cohesive in recent Blytheville School Board elections.
Racial polarization, however, is the essential component of the third Gingles precondition — a pattern of white bloc voting that permits the majority usually to defeat the minority’s preferred candidates, thereby diluting the minority’s vote. Racial polarization requires a focus on whether “black voters and white voters vote differently.” Gingles, 478 U.S. at 53 n. 21, 106 S.Ct. at 2768 n. 21; Clay, 896 F.Supp. at 935-36. In making findings of racial polarization, “we rely primarily on actual events and practical politics.” Jeffers v. Clinton, 730 F.Supp. 196, 208 (E.D.Ark.1989), aff'd, 498 U.S. 1019, 111 S.Ct. 662, 112 L.Ed.2d 656 (1991). Therefore, in measuring this factor, elections under a prior electoral system, elections in which there was no African-American candidate, the size and influence of the white crossover vote, and the strength of a minority preferred candidate’s support become relevant. See Niagara Falls, 65 F.3d at 1012-17; Southern Christian Leadership Conference v. Sessions, 56 F.3d 1281, 1293 (11th Cir.1995) (plaintiffs’ expert’s analysis flawed because he only analyzed elections involving a minority candidate), cert. denied, — U.S. -, 116 S.Ct. 704, — L.Ed.2d - (1996). For the most part, plaintiffs did not gather such data, and what there is in the record they urge us to ignore.
The district court’s finding of “racially polarized” voting reflected Lynch’s analytical error:
The testimony of Mr. James Lynch at trial, based on analyzing past elections and voting patterns, established that voting in the School District is racially polarized in that usually the majority of black voters vote for black candidates and the majority of white voters vote for white candidates.
Harvell, 759 F.Supp. at 527-28 (emphasis added). This is simply a mislabeled finding of political cohesiveness, the second Gingles precondition. This court then converts that limited finding into a far broader finding of “highly polarized” voting. A brief review of the undisputed facts demonstrates that the record -will not support this additional finding:
• From 1975 until 1991, two of the eight School Board members were African-Americans. Until Norvell Moore, a four-term incumbent, elected not to run in 1991, the 29% African-American voting age minority had succeeded in electing 25% of the School Board for sixteen straight years, under an entirely at-large election system. Presumptively, therefore, racially polarized white bloc *1393voting has not unlawfully diluted the minority’s vote.
• In 1982, two African-American candidates each ran against a single white opponent. Dr. Helen Nunn won, receiving 33% of the votes in precincts having more than an 80% white voting age population. Incumbent Ayre Lester received 19% of the votes in those precincts and lost. In 1987, Norvell Moore ran head-to-head against a white opponent. Mr. Moore received 46% of the votes in those predominantly white precincts and won. This is undeniable evidence of legally significant, continuing white crossover voting. Yet it was not assessed by the district court because Lynch did not know the difference between racially cohesive voting and racially polarized voting. This court then compounds that legal error with the question-begging observation that white crossover voting cannot be legally significant because there has been a finding of “consistent polarization.” Ante at 1387.
• Plaintiffs’ analysis relies on the fact that African-American candidate Shirley Harvell received 79% of the votes cast in 1990 in the predominantly minority Robinson Elementary School voting district. But Harvell received only 63 votes from that district, whereas Dr. Nunn received 247 votes from that district in her 1982 victory, and Mr. Lester, who barely lost in 1982 despite being in very poor health, received 227 votes at Robinson Elementary School. Without knowing the size of the voting age population of each district, as well as its percentage of minority voters, we cannot assess whether white bloc voting was the likely cause of Harvell’s defeat.
The Supreme Court recently cautioned that “ ‘minority political cohesion’ and ‘majority bloc voting’ showings are needed to establish that the challenged districting thwarts a distinctive minority vote by submerging it in a larger white voting population. Unless these points are established, there neither has been a wrong nor can be a remedy.” Growe v. Emison, 507 U.S. 25, 40-41, 113 S.Ct. 1075, 1084, 122 L.Ed.2d 388 (1993) (citation and footnote omitted). This court has improperly transformed the district court’s finding that most voters vote for candidates of their own race into a blanket assumption that racially polarized white bloc voting usually prevents the election of minority preferred candidates. The record will not support that appellate court finding. Compare Magnolia Bar Ass’n, Inc. v. Lee, 994 F.2d 1143, 1148-50 (5th Cir.) (where African-American candidates won two high profile elections with 58% and 30% of the white vote, no. clear error in finding the third precondition not satisfied despite expert’s testimony that white bloc voting was legally significant and these elections were aberrational), cert. denied, - U.S. -, 114 S.Ct. 555, 126 L.Ed.2d 456 (1993).
B. Minority Preferred Candidates Are Not Usually Defeated
The third Gingles precondition also requires proof that white bloc voting usually will defeat “the minority’s preferred candidate.” 2 The court finds that all losing African-American candidates since 1987 were minority preferred, looking only at whether they received most of the votes in minority-dominated districts. That is a logical assumption in most cases involving a racially cohesive electorate. But we are reviewing the district court’s contrary finding under a clearly erroneous standard that “preserves the benefit of the trial court’s particular familiarity with the indigenous political reality.” Gingles, 478 U.S. at 79, 106 S.Ct. at 2781. In conducting that review, I begin with what the district court actually found:
[P]rior to 1988, the lowest number of votes that any black candidate received was 629 and the highest was 1232. On the average, black candidates received nearly 977 votes. On the other hand, since 1988, the most votes received by any black candidate was 374 and the lowest was 135. The average since 1988 has been 237 votes per black
*1394candidate. White candidates received an average total of 1264 votes prior to 1988, with the highest being 2212 and the lowest being 606. Since 1988, white candidates have received an average of 1064 votes, with the highest being 1345 and the lowest being 758. The following chart of averages is helpful.
Black Candidates White Candidates
1969-1987 1264 977
1988-1064 237
Percent Change 15.8% 75.7%
... The Court is of the opinion that none of the black candidates since 1988 have been the “preferred” candidate of the black community.... Plaintiffs have failed to show that the reason they are unable to maintain proportional representation is because of the way the white majority votes.... Indeed, one of the candidates who had been the overwhelming choice of the black population [Norvell Moore] chose not to run for reelection.
A review of the entire record persuades me that this finding is not clearly erroneous:
• If the eight unsuccessful African-American candidates since 1987 had received as many votes as Dr. Nunn received in her successful contested race in 1982 (the first time she ran), three would have won outright and two more would have forced run-off elections under the new majority-vote rule. (Given the results of prior elections, the court’s speculation that whites will always win run-off elections, ante at 1388-89, is unwarranted.)
• In 1988, African-American candidate Curtis Smith lost an election to white candidate Bill Sullivan, 758 to 166. Smith did not carry the Mississippi County Implement Company district, which has a BVAP of 80%, by far the highest of any voting district. Yet the court finds that Smith was the minority preferred candidate.
• Plaintiffs’ Exhibit 26, a copy of the official election returns, reports that in 1990, white candidate Littrell received 62 votes and plaintiff Harvell received 7 votes in the East End Fire Station district, one of three districts having a higher BVAP than the Blythe-ville School District as a whole. Plaintiffs’ Exhibit 14, prepared by Lynch, reports that plaintiff Harvell captured 60% of the vote in that district, rather than the 10% she actually received. Based upon Lynch’s grievous error, plaintiffs contend and now this court finds that Harvell was the minority preferred candidate in that election.
• The votes cast in the two districts with a BVAP majority accounted for 26% of the total votes in the 1982 election (when Dr. Nunn won and Mr. Lester barely lost), but only 14% of the votes in 1989, 12% in 1990, and 13% in 1991.
• Only one African-American candidate since 1987 has received a majority of the votes east in the district that most closely mirrors the total School District population (40% minority). In 1982, Dr. Nunn and Mr. Lester received over 75% of the votes in a similar district.
In my view, this is overwhelming evidence that factors other than racially polarized voting, or the 1987 election law change, account for the election defeats of African-American candidates in recent years. Yet the court dismisses the obvious import of this evidence with the comment that it “stands to reason” that the 1987 change in the law caused the precipitous drop in minority voter turnout. Ante at 1388. I disagree. Prior to 1987, African-American candidates had won four of the previous five contested elections against white candidates. They had received an absolute majority in two of the previous three elections. With that voting history, only preconceived notions of racial voting behavior could lead the court to conclude that the majority-vote rule caused minority voters to lose all hope of success.
We should not forget that we deal here with the politics of education, not with the election of general purpose legislators. Dr. Nunn testified on cross examination that she opposed single-member districts “[b]ecause I have spent my life working with the total *1395community, working together, and I think that’s one way to improve quality of education.” Plaintiff Hattie Middlebrook, an unsuccessful candidate in 1990, testified that Board members Norvell Moore and Dr. Nunn were no longer minority preferred candidates because of their consensus-building political views. On this record, I reject the court’s unsupported speculation that it was the spectre of white bloc voting, or the change to majority rule in 1987, that caused the vast majority of potential African-American voters not to support candidates such as Ms. Middlebrook. The district court’s findings as to the third Gingles precondition should be affirmed.
C. Consistent and Sustained Minority Representation
In Gingles, six Justices agreed that “consistent and sustained success by candidates preferred by minority voters is presumptively inconsistent with the existence of a § 2 violation.” 478 U.S. at 102 (O’Connor, J., concurring). While proportional minority representation is not a safe harbor that automatically defeats a claim of vote dilution, it should be given “extremely heavy weight.” African Am. Voting Rights Legal Defense Fund, Inc. v. Villa, 54 F.Sd 1345, 1355-56 (8th Cir.1995). Section 2 does not “require maximization of minority-group representation.” Little Rock Sch. Dist. v. Pulaski County Special Sch. Dist., 56 F.3d 904, 910 (8th Cir.1995) (no § 2 violation where minority comprising 29% of the district’s voting age population has held 28.6% of the Little Rock School Board seats continuously for twenty years).
In this case, an African-American has held a seat on the School Board every year since 1974, and there were two African-Americans on the Board — proportional representation— for sixteen consecutive years. The court dismisses this “consistent and sustained success” as the product of “special circumstances.” For example, the court disregards Dr. Nunn’s most recent success because she ran unopposed for a third term in 1990. While Gingles noted that incumbency and lack of opposition may be “special circumstances” that do not disprove a racially polarized electorate, a finding to that effect requires factual analysis. Dr. Nunn’s prior victories and standing in the at-large community may well have persuaded potential white candidates that she was unbeatable. In a district with only 29% minority voters, that would be strong evidence that the electorate is not so racially polarized that African-Americans “have less opportunity ... to elect representatives of their choice,” the ultimate issue under § 2. See Sanchez v. Bond, 875 F.2d 1488, 1493 (10th Cir.1989), cert. denied, 498 U.S. 937, 111 S.Ct. 340, 112 L.Ed.2d 305 (1990). “It borders on the absurd, of course, to suggest that an individual’s success in politics should be discounted as aberrational because the person is qualified and popular.” Niagara Falls, 65 F.3d at 1021 n. 22.
I am particularly distressed by the court’s suggestion that the challenged system violates § 2 because it “allows only for victory of majority-preferred minority candidates,” ante at 1389, and its criticism of the manner in which successful African-American candidates “managed to obtain and retain their seats,” ante at 1390. Apparently, the court believes that only a racially balkanized election system can comply with the Voting Rights Act. I subscribe to a contrary principle — that “[n]o legal rule should presuppose the inevitability of electoral apartheid — least of all a rule interpreting a statute designed to implement the Fourteenth and Fifteenth Amendments to the Constitution.” Niagara Falls, 65 F.3d at 1016.
D. Conclusion
For the foregoing reasons, I conclude that plaintiffs have failed to prove that legally significant white bloc voting in the Blythe-ville School District usually defeats a minority preferred candidate. Accordingly, they have not satisfied the third Gingles precondition, nor have they established by a totality of the circumstances that the challenged election system results in the minority vote dilution that § 2 prohibits.
I do not know whether the 1987 majority-vote law will ultimately work to deprive minority voters in the Blytheville School District of their rights under § 2. It may be *1396that minority-preferred candidates will emerge but persistently fail to achieve proportional electoral success. In that case, § 2 relief will no doubt be warranted because, in the presence of strongly polarized white bloc voting, a majority-vote rule combined with at-large voting districts and a 30% minority voting age population most likely violates § 2. Cf. City of Port Arthur v. United States, 459 U.S. 159, 167, 103 S.Ct. 530, 535, 74 L.Ed.2d 334 (1982). But absent proof of sufficiently polarized white bloc voting, § 2 relief should be denied.3 Our function under the Voting Rights Act is not “to dictate to the provinces the ‘correct’ theories of democratic representation, the ‘best’ electoral systems for securing truly ‘representative’ government, the ‘fairest’ proportions of minority political influence, or ... the ‘proper’ sizes for local governing bodies.” Holder v. Hall, — U.S. -, -, 114 S.Ct. 2581, 2602, 129 L.Ed.2d 687 (1994) (Thomas, J., concurring). Therefore, I would affirm.
APPENDIX I
The following table is a compilation of election results for the Blytheville, Arkansas, School Board from 1969 to 1992. Black candidates are designated by “B” and white candidates are designated by “W.”
[[Image here]]
*1397[[Image here]]
APPENDIX II
YEAR CANDIDATES CORRELATION COEFFICIENT*
1988 Bill Sullivan Curtis “Preacher” Smith 0.8437
1989 Harold Sudbury, Jr. Thurman J. Green, II 0.9582
Steve Littrell Shirley M. Harvell 0.9674
Steward R. Jerome Lawrence B. Haley 0.9678
1990 Steve Littrell Shirley M. Harvell 0.9086
*1398[[Image here]]
. In addition, Lynch was not a properly qualified expert. He disclaimed expertise in statistics. He described the correlation coefficient as “simply a technique which one can employ.” He could not explain his use of the "R square” factor, and he did not even attempt to explain the "F value" by which he purported to find statistical significance. It is doubtful that his statistical analysis and opinions were even admissible under Fed.R.Ev. 702.
. By submitting data only on elections in which there was an African-American candidate, plaintiffs improperly shifted the Gingles focus from “minority preferred candidates” to "minority candidates.” In my view, given the sustained success of many African-American candidates in the recent past, a finding of racially polarized white bloc voting could only be made after thorough analysis of the relative success of the minority preferred candidates in all elections.
. If plaintiffs were entitled to relief, it would only be an injunction against the 1987 majority-vote rule. The court's opinion implies that the district court must now enter an "appropriate remedial decree” creating single-member voting districts. Given the sustained success of minority preferred candidates under the prior, plurality at-large system, this remedy violates the congressional intent that the 1982 amendments to § 2 not be construed as "an all-out assault on at-large election systems in general.” S.Rep. No. 417, 97th Cong., 2d Sess. 27 (1982), reprinted in 1982 U.S.C.C.A.N. 177, 205. Thus, I also dissent from the remedial portion of the court’s decision.
The Correlation Coefficient (the “r” statistic) measures the strength of a relationship between two variables. The "r" may range from 0.0 (indicating the two variables are independent) to +1.0 (indicating the two variables are perfectly correlated in a positive direction).