Rodriguez v. Bexar County Texas

EDITH H. JONES, Circuit Judge: This case arises out of Bexar County’s redistricting of its Justice of the Peace and Constable Precincts following the 2000 national census. The plaintiffs contended that the redistricting plan impermissibly dilutes the votes of Hispanics in violation of Section 2 of the Voting Rights Act and that it violates Article V, Section 18 of the Texas Constitution. The district court ruled in favor of the plaintiffs on then-federal law claim and denied relief on the state constitutional claim. After carefully reviewing the evidence adduced at trial, we hold that there is no legal or factual basis for the court’s finding of vote dilution and that the plaintiffs are entitled to no relief under federal law. We reverse in part, affirm in part and vacate the district court’s injunctive relief. I. BACKGROUND In August 2001, Bexar County’s Commissioners Court adopted, pursuant to the Texas Constitution, a redistricting plan for its Justice of the Peace and Constable Precincts (“Justice Precincts”). See Tex. Const, art. V, § 18. The major changes effected by the 2001 redistricting plan were the reduction of the number of precincts from five to four, and the elimination of one constable position. The plan eliminated Precinct Five, which elected one Justice of the Peace and one Constable, but it added one new Justice of the Peace position to Precinct One, thereby leaving the total number of Justice of the Peace positions unchanged. Constable Tejeda’s Precinct Five position was eliminated. Under both the current and for *859 mer redistricting plans, there are one majority-black and two majority-Hispanic districts. The' population of now-extinct Precinct Five was transferred into revised Precincts One and Two. The redistricting plan was pre-cleared by the Department of Justice’s Civil Rights Division. Just after elections had been held under the new plan, the plaintiffs filed suit against Bexar County alleging that the plan violated Sections 2 and 5 of the Voting Rights Act and Aetiole V, SeotioN 18 of the Texas Constitution. 1 The district court conducted a bench trial and ruled in favor of the plaintiffs on their Section 2 vote dilution claim; As a remedy, the court ordered the results of the 2002 elections set aside, 2 and the judge reinstated the original five-precinct plan, and, inter alia, ordered Bexar County to re-fund Constable Tejeda’s post. This court stayed the court’s remedy pending Bexar County’s appeal. II. DISCUSSION A. Voting Rights Act Claim What the plaintiffs precisely assert is that the elimination of Precinct Five and its consolidation in the other redrawn districts has diluted the influence of Hispanic votes in Precinct Two. It is -surely no accident, however, that former Constable Tejeda, whose position was eliminated in the redistricting, is the lead plaintiff. Section 2 of the Voting Rights Act proscribes vote dilution whereby a class of citizens has “less opportunity than other members of the electorate to participate in the political process and to elect representatives of their choice.” 42 U.S.C. § 1973. This court applies a two-step framework in analyzing Section 2 claims. NAACP v. Fordice, 252 F.3d 361, 365 (5th Cir.2001). First, plaintiffs challenging a redistricting plan must satisfy the preconditions for a Section 2 claim set forth by the Supreme Court in Thornburg v. Gingles, 478 U.S. 30, 106 S.Ct. 2752, 92 L.Ed.2d 25 (1986). 3 Id. Second, the plaintiffs must prove- that based on the “totality of the circumstances,” the challenged plan results- in the denial of the right to vote based on color or race in violation of Section 2. Fordice, 252 F.3d at 366. To meet the threshold Gingles test, the plaintiffs bear the burden to show, -by a preponderance of the evidence, that: (1) the affected minority group is sufficiently large and geographically compact to constitute a voting age majority in a district; (2) the minority group is politically cohesive; and (3) the majority votes sufficiently as a bloc that it is able—in the *860 absence of special circumstances—usually to defeat the minority group’s preferred candidate. Id. (citing Gingles, 478 U.S. at 50-51, 106 S.Ct. at 2766-67). In reviewing a district court’s decision regarding an alleged violation of Section 2 of the Voting Rights Act, this court analyzes the legal standards applied by a district court de novo, id. at 364, and the factual findings for clear error. Gin-gles emphasized that the proper assessment of vote dilution claims is “peculiarly dependent upon the facts of each case” and requires “an intensely local appraisal of the design and impact of the contested electoral mechanisms.” 478 U.S. at 79, 106 S.Ct. at 2781. The clear error standard precludes reversal of a district court’s findings unless we are “left with the definite and firm conviction that a mistake has been committed.” Anderson v. City of Bessemer, 470 U.S. 564, 573, 105 S.Ct. 1504, 1511, 84 L.Ed.2d 518 (1985). We may not reverse for clear error so long as the district court’s findings are “based on a plausible account of the evidence considered against the entirety of the record.” Fordice, 252 F.3d at 365. The parties do not dispute that the plaintiffs satisfy the first two prongs of the Gingles threshold inquiry: Hispanics are sufficiently numerous and geographically compact to constitute a voting age majority in Justice Precinct Two, and they are politically cohesive. The evidence adduced before the district court focused on Gin-gles’ third inquiry, i.e., the ability of Hispanics to elect their preferred candidate under the 2001 plan in reapportioned Justice Precinct Two. 4 The plaintiffs’ argument is that although Precinct Two retains a majority of Hispanic residents, the majority is narrower than that in former Precinct Five and, having been diluted, is barely sufficient to ensure Hispanic electoral success. Since neither party presented significant evidence regarding the other redistricted precincts, our analysis is confined to Justice Precinct Two. The critical question before the district court, and now on appeal, is whether the plaintiffs met their burden of proof on the third Gingles factor. Lacking such proof, the plaintiffs cannot succeed. See, e.g., Magnolia Bar Ass’n, Inc. v. Lee, 994 F.2d 1143, 1148 (5th Cir.1993). After carefully reviewing the record, and being cognizant of the deference owed to the district court, we have concluded that the district court made substantial legal and factual errors in evaluating the plaintiffs’ evidence. Principally, but not solely, the court erred in ignoring the defendants’ reconstituted election analysis, and it erred in applying the “special circumstances” test to ignore the consistent electoral victories of Hispanic candidates in Precinct Two. As a result, the district court clearly erred in ultimately concluding that the 2001 redistricting plan impermissibly diluted the Hispanic vote in Bexar County. 1. Reconstituted Election Analysis Because, at the time of trial, only one election had been held within the new precinct boundaries created by the 2001 plan—the 2002 election for Bexar County Constable in Justice Precinct Two—experts for both sides agreed that it was appropriate to look to exogenous races to determine whether racial bloc voting took place in the revised Justice Precinct Two. 5 *861 In doing so, the experts employed reconstituted election analysis to evaluate the results of 12 exogenous races for other elected posts that took place in 2002. 6 Both experts agreed that these races were the most relevant to determining whether Anglos voted as a bloc usually to defeat the Hispanic candidate of choice in Precinct Two. 7 In addition, the plaintiffs’ expert responded to the defendant’s expert’s analysis of other prior election cycles and agreed that the results from the 2000 elections also had substantial probative value. Not surprisingly, the expert opinions conflicted on the critical point. Compare Ex. D-9 with Exs. P-108, P-146. The district court, noting these contrary conclusions, determined that the reconstituted election analyses presented by both sidés were insufficient and unpersuasive, and opted to consider other evidence to make its determination as to the third Cingles factor. The district court discarded the reconstituted election evidence offered by both parties for two reasons. First, the court found the reconstituted election methodology to be inherently unreliable because including or excluding what the district court believed were a “handful” of “over and under” ballots could lead to substantially different conclusions on the ultimate question of racial bloc voting. In addition, the court found that “special circumstances” in both the 2000 and 2002 election cycles made these elections unreliable for the purpose of evaluating the validity of the 2001 redistricting plan. Unfortunately, the district court clearly erred in both determinations. Reconstituted election analysis is a relatively simple method that extracts actual election results from a variety of statewide and local races that subsume the area being analyzed and determines, precinct-by-precinct within the new district, the racial composition of the yote and the “winner” within the new district. This method of aggregation allows a researcher to determine how an individual candidate performed within the boundaries of the target district even though the actual election covered a different geographical area. The defendant’s expert, Dr. John Alford, employed this-standard method in examining 13 races from the 2002 general election. The plaintiffs’ expert, Dr. Henry Flores, also employed a similar method—with one crucial difference: in calculating the percentage of the vote received by each candidate, Dr. Flores used the correct numerator—the total votes cast for each candidate within the boundaries of Justice Precinct Two. However, in calculating the appropriate denominator, Dr. Flores did not use the total votes cast in each race within Precinct Two, but rather used the total ballots cast in Precinct Two in the overall election. This approach systematically misrepresents the percentage of the vote obtained by each candidate. It does so by improperly including “over-votes” and “under-votes” in *862 the denominator of the equation. Over-votes are ballots where a voter casts more than one vote for an office and thus invalidates his vote for that office. Under-votes are those ballots where a voter does not mark any candidate for a given office. Neither over-votes nor under-votes “count” toward determining victory in a race because they represent, respectively, either a “spoiled ballot” or an uncast ballot for that particular race. The ballot remains valid, of course, for those races and only those races in which it was properly marked. But to use such ballots in the denominator for calculating the percentage of the vote received in a given race, when the votes from those ballots would not have affected the race, is simply incorrect and results in skewed and inaccurate vote percentages. Dr. Flores’s report erred in this manner. Based on the inaccurate percentages he calculated, Dr. Flores concluded that only eight of the 13 Hispanic candidates of choice “won” their races within the boundaries of Justice Precinct Two. 8 At trial, Bexar County’s counsel carefully dissected this error, leading Dr. Flores to admit, contrary to his written report, that the proper calculation would have shown that 12 out of 13 Hispanic candidates of choice actually “won” the exogenous elections within the confines of Justice Precinct Two. 9 ROA vol. 11 at 81. To its credit, the district court recognized this fatal flaw in Dr. Flores’s methodology. However, instead of simply discarding Dr. Flores’s flawed findings and relying on the proper calculations made by the defendant’s expert, the district court held that Dr. Flores’s errors demonstrated “how easily reconstituted election analysis can be abused under the best of circumstances.” As a result, the district court found the “evidence generated from these reconstituted election analyses to be largely unpersuasive.” The district court then inexplicably threw out the defendant’s expert evidence based on the flawed methodology employed by the plaintiffs’ expert. The district court first erred in suggesting that the difference between the two expert analyses arose from the inclusion of a “handful of ballots.” To take just one example, in the Garza-Wilborn race for Justice Precinct Two Constable, Dr. Flores used the total ballots cast in the election— 83,968 — as the denominator in his equation to determine that Garza received 44.6 per *863 cent of the vote. See ROA vol. 11 at 76. 10 However, Dr. Flores’s methodology improperly included 6,372 under-votes and 24 over-votes from that election. Id. These votes, taken together, constitute 7.6 percent of the ballots used in the denominator of Dr. Flores’s equation. Omitting .these uncounted ballots from the calculation, as should have been done, reveals that Garza actually obtained 48.3 percent of the vote. See Ex. D-9. Thus, we disagree that the mistakenly included ballots were a mere “handful.” On the contrary, Dr. Flores’s calculation errors were significant and systematic, and produced a substantially flawed analysis. These erroneous results do not cast doubt on the methodology of reconstituted election analysis, but on the quality of the particular calculations. 11 The district court, though appropriately disturbed at the serious errors made by the plaintiffs, chose to toss out all of the reconstituted election evidence put forward by both sides. But as the defendant’s brief points out, such an approach is similar to determining that mathematics is a flawed science simply because one expert testifies that two plus two is four and another expert testifies that two plus two is five.. The court should have considered Dr. Flores’s errors as undermining the weight of his testimony, not that of the defendant’s expert. See Rollins v. Fort Bend Indep. Sch. Dist., 89 F.3d 1205, 1219 (5th Cir.1996) (holding that numerous errors by an expert witness can make all of that expert’s findings and theories unreliable). An independent assessment of the validity of the defendant’s expert testimony would then have shown that analysis completely uncontradicted in its findings that the Hispanic candidate of choice obtained more votes within new Justice Precinct Two in 12 out of the 13 races identified by both experts as central to the court’s Gingles analysis. Moreover, when the 2000 general election is included in this analysis, the defendant’s uncontradicted expert testimony indicates that the Hispanic candidate of choice would have won 21 of the 22 most recent races within Justice Precinct Two. These results contradict the district court’s determination that the Anglo bloc voting serves usually to defeat the Hispanic candidate of choice in Justice Precinct Two. The district court’s decision to disregard these results constitutes clear error. This error prevented the district court from considering the data that both sides agreed were the most probative on the third Gingles factor, and thus, strikes at the core of the district court’s ultimate conclusions. 2. Evaluating “Special Circumstances” In its initial opinion, the district court acknowledged that both sides focused on *864 expert testimony regarding the 2000 and 2002 elections and concurred on the importance of these elections in proving the legality of the 2001 redistricting plan. Nevertheless, the court held that these elections exhibited “special circumstances” that made “an accurate extrapolation of the redistricting plan’s effect ... impossible.” The district court identified as the relevant “special circumstances” the presence of a Hispanic candidate, Tony Sanchez, a Democrat running for Governor in 2002, and the presence of George W. Bush, at the time the Republican Governor of Texas, as a candidate for President in 2000. The Supreme Court has cautioned that “special circumstances ... may explain minority electoral success in [an otherwise] polarized contest,” and that such aberrational victories do not necessarily disprove racial vote dilution. Gingles, 478 U.S. at 57, 106 S.Ct. at 2770. The district court, however, misapplied the special circumstances analysis in a manner contrary to that contemplated by the Supreme Court and this circuit’s precedents. As explained in Gingles, the special circumstances analysis was designed to prevent defendant jurisdictions from arguing that a minority candidate’s occasional victory in an otherwise racially polarized electorate defeats a vote dilution claim. Id. To this end, the Court listed several factors that might contribute to the unusual success of an individual minority candidate—the absence of an opponent, incumbency, or utilization of “bullet voting” procedures. Id. While not exhaustive, this list comprises circumstances that might explain a victory for a minority candidate in a polarized district. This circuit accordingly holds that while special circumstances may be used to “explain a single minority candidate’s victory,” the Supreme Court’s comment regarding such circumstances “cannot be transformed into a legal standard which requires the court to force each and every victory of several minority candidates to fit within a prescribed special circumstance.” Rollins, 89 F.3d at 1213 (emphases added). The Rollins court went on to note that “[e]very victory [of a minority candidate] cannot be explained away as a fortuitous event.” Id. In the present case, the district court employed the Gingles special circumstances analysis not to explain the victory of an individual minority candidate, but rather to explain away the consistent success of Hispanic candidates in a number of races over two general election cycles. This was clear error. As noted above, reconstituted election analysis—performed with the proper arithmetic—demonstrates that the Hispanic candidate of choice won 21 of 22 contests during the 2000 and 2002 general elections within Justice Precinct Two. Moreover, even if “special circumstances” could be used to explain away all of the minority candidate victories, the evidence fails to provide any basis for such a finding in this case. The district court reasoned that in the 2000 elections, George W. Bush’s candidacy for President likely affected both Hispanic and overall voter turnout, but the court failed to explain what the “special circumstances” of Bush’s candidacy might have been. Without evidence, it is impossible to tell whether Bush’s candidacy helped or harmed Hispanic-favored candidates in 2000. We do not.doubt that Bush’s candidacy had some effect on turnout within Texas. The plaintiffs’ expert believed that Hispanic election success was more difficult in 2000. However, because no other evidence described the magnitude or nature of this effect, the court clearly erred in speculating how Bush’s candidacy explained the over *865 whelming success of Hispanic-favored candidates. 12 Similarly, the district court clearly erred in determining that the presence of Tony Sanchez at the top of the ticket in 2002 was a special circumstance that explained the success of Hispanic candidates in that election cycle. No evidence presented at trial tended to indicate that Hispanic candidates were more likely to succeed as a result of Sanchez’s candidacy. Instead, the plaintiffs’ expert testified that he had not conducted a study regarding the effect of Sanchez’s candidacy, or what he termed “racist” anti-Sanchez ads, on Hispanic turnout, and he opined that there was “no way of telling the effects of how those ads played out.” ROA vol. 11 at 135-36. Indeed, the plaintiffs’ expert went on to testify specifically that Sanchez’s candidacy and the related ads “could have increased turnout on both sides.” Id. at 136. The lack of supporting evidence wholly undermines the district court’s finding that these elections were not sufficiently reliable to provide any insight into the polarized voting inquiry. The district court’s finding in this regard is clearly erroneous. Because they are not vulnerable to a special circumstances attack and were not otherwise disputed, the 2000 and 2002 election results, as properly reconstituted, have substantial probative value on the question whether the plaintiffs met the third Gingles precondition. The evidence of overwhelming electoral success for Hispanic-favored candidates over a wide range of offices and in two separate general election cycles, in both a Presidential and a non-Presidential election year, leads to the firm and definite conclusion that the district court clearly erred in finding that Anglos vote as a bloc usually to defeat the Hispanic candidate of choice within new Justice Precinct Two. Recent voting patterns and trends suggest that Hispanics will continue to enjoy substantial success in electing the candidates they support in Justice Precinct Two. 13 3. Other Statistical Evidence After erroneously discarding the probative reconstituted election analyses, the district court purported to rely on the “homogenous precinct analysis” conducted by the plaintiffs’ expert. At trial, the plaintiffs submitted expert reports and testimony developed using both regression and homogenous precinct analysis. See, e.g., ROA vol. 11 at 22-24. The homogenous precinct analysis conducted by the plaintiffs’ expert selected predominantly *866 Hispanic and Anglo precincts 14 from within Justice Precinct Two and examined the performance of the 13 candidates who ran in the 2002 elections. Id. We need not consider this homogeneous precinct analysis further, however, because, despite its statements to the contrary, the district court did not actually rely on this analysis. 15 The court’s opinion makes clear that it confused the plaintiffs’ homogenous precinct analysis with the separate analysis conducted by the plaintiffs’ expert of some 115 jurisdictions 16 that had been redistricted on a single member concept and which contained varying percentages of Hispanic voters. 17 These election jurisdictions consisted of San Antonio city council districts, state house and senate districts, state board of education districts, and U.S. Congressional districts. In each of these election jurisdictions, the plaintiffs’ expert examined the overall population and voting age population by racial group, as well as the “Spanish-surname” 18 voter registration. See ROA vol. 11 at 25-26; Pis. Ex. 14-15. In addition, the plaintiffs’ expert focused on the success rates of Hispanic candidates who ran for election in *867 these districts. See ROA vol. II at 25-52. The expert’s analysis of single member districts is completely separate and distinct from the homogenous precinct analysis. Based on his examination of the success rates of Hispanic candidates in Bexar County, the plaintiffs’ expert confirmed, at least in his mind, the conclusion he had reached based on his earlier homogenous precinct analysis that examined the results of the 13 reconstituted elections from the 2002 general election. Dr. Flores concluded that, for a Hispanic candidate to succeed in Bexar County, the “Spanish-surname” registered voter population must exceed 50 percent in a given election jurisdiction. Accepting this conclusion, the court held that because new Justice Precinct Two [does] not contain a percentage of registered voters in excess of 50 percent ... the Court is in agreement with Dr. Flores that Anglo voters, in the absence of special circumstances, can and will vote as a. bloc in new Precinct Two usually to defeat the candidate chosen by Hispanics ... [and therefore] the third prong of Gingles is satisfied. The district court’s determination in this regard is clearly erroneous for two reasons. First, we know of no caselaw that simply correlates minority candidate success rates, absent any additional statistical analysis, with a minimum threshold of minority voter registration that automatically satisfies Gingles’ third prong. Indeed, in our view, such an approach cuts at the heart of Gingles and its progeny, which prohibit courts from presuming racial bloc voting and require the plaintiffs to prove that Anglos actually vote as a bloc usually to defeat the minority candidate .of choice. See Growe v. Emison, 507 U.S. 25, 41, 113 S.Ct. 1075, 1085, 122 L.Ed.2d 388 (1993) (“a court may not presume bloc voting even within a single minority group”) (citing Gingles, 478 U.S. at 46, 106 S.Ct. at 2764). Because the district court discounted the reconstituted election evidence submitted by both parties, 19 it had no information that would have shed any light on whether Anglo voters in these areas actually vote as a bloc usually to defeat the Hispanic candidate of choice. Rather,. in relying only upon the remaining data submitted by the plaintiffs in support of their expert’s conclusion, the district court im-permissibly presumed Anglo bloc voting against Hispanic candidates in any and all districts where Hispanic voter registration is below 50 percent. Moreover, even if this method of analysis were appropriate, the data relied upon by the district court are ambiguous at best on the question whether 50 percent is the minimum threshold for Hispanic voter registration in order to assess Section 2 compliance. The data shed little, if any, light on the real question in this case, i.e., whether the 48 to 49 percent 20 of Hispanic voters in Justice Precinct Two is sufficient that they have an opportunity to elect the *868 candidate of their choice. As Bexar County points out, of the 115 districts examined by Dr. Flores, 71 have over 50 percent Hispanic registered voters. See Exs. P-14, P-15. It is undisputed that when the percentage of Hispanic registered voters exceeds 50 percent, Hispanics have a clear opportunity to elect their candidates of choice in those districts. But, the question before the district court was whether a lower percentage of registered voters was sufficient to afford such an opportunity. Of the remaining 44 districts, 41 had less than 32 percent Hispanic registration and 26 of these had less than 20 percent Hispanic registration. Id. Evidence regarding such heavily non-Hispanic districts does not carry the plaintiffs’ burden as to Precinct Two. In fact, in the only relevant set of three districts where Hispanic voter registration lay between 48 and 49 percent, a Hispanic candidate was actually elected. 21 While the experts disagreed as to the reasons for the success of the Hispanic candidate in this below-50 percent district, the results undermine Dr. Flores’s finding that 50 percent Hispanic voter registration is a “magical number” below which the third Gingles factor may be presumed to be satisfied. 22 For the foregoing reasons, the only supporting evidence referenced by the district court on the third Gingles prong was actually non-probative, and the court’s finding was therefore clearly erroneous. 23 4. Plaintiffs’Burden of Proof A final observation is in order concerning plaintiffs’ burden of proof of the third *869 Gingles factor. Elections for the three offices in Justice Precinct Two — one constable and two justices of the peace — are held in November of even-numbered years, with two positions on the ballot in presidential election years and one position on the ballot in non-presidential election years. On cross-examination, Dr. Flores admitted that he agreed with the defense expert, Dr. Alford, that in presidential election years, Anglos do not vote as a bloc usually to defeat the Hispanic candidate of choice in Justice Precinct Two. See ROA vol. 11 at 65-67. In addition, Dr. Flores admitted that it was unclear and uncertain whether Anglos vote as a bloc, in non-presidential election years', usually to defeat the Hispanic candidate of choice within the challenged area. Id. at 67-68. 24 Given that the plaintiffs’ own expert witness at trial admitted that for two out of the three relevant election scenarios the third Gingles factor could not be met, and that in the remaining election scenario, it is unclear and uncertain whether this factor could be met, the district court’s finding that the plaintiffs had satisfied their burden of proof on this factor is all the more perplexing. 5. Totality of the Circumstances Inquiry Because the plaintiffs failed to meet their burden of proof on one of the three essential Gingles preconditions for a Section 2 vote dilution claim, we need not reach the County’s claim that the district court clearly erred in its analysis of the totality of the circumstances. Magnolia Bar, 994 F.2d. at 1148. Nonetheless, where plaintiffs are able to satisfy the Gingles threshold inquiry and a district court properly turns to the requisite totality of the circumstances analysis, district courts must thoughtfully consider the factors enumerated in Zimmer v. McKeithen, 485 F.2d 1297, 1305 (5th Cir.1973), and the Senate Judiciary Committee report on the 1982 amendments to the Voting Rights Act, S.Rep. No. 417, 97th Cong., 2d Sess. 28-29 (1982), reprinted in 1982 U.S.C.C.A.N. 177, 206-07. The district court’s relatively cursory analysis of the factors was insufficient in this case. In particular, the court ignored that five of the 11 officials elected from the county’s justice precincts are Hispanic and that as the election cycle progresses in the new Precinct Two, more Hispanics will likely be elected. 25 This powerful evidence of nearly proportional representation ought to have been considered by the court. 6. Relief Ordered by the District Court From the preceding discussion, it is clear that the. district, court’s award of injunctive relief must be vacated. We pause briefly, however, to note that district courts should use a great *870 deal of caution in invalidating the results of a duly held election and ordering the implementation of its own alternative dis-tricting plan. The primary responsibility for correcting Voting Rights Act deficiencies rests with the relevant legislative body. Jones v. Lubbock, 727 F.2d 364, 387 (5th Cir.1984), reh’g denied, 730 F.2d 233 (1984) (citing Chapman v. Meier, 420 U.S. 1, 27, 95 S.Ct. 751, 766, 42 L.Ed.2d 766 (1975)). Both the Supreme Court and this court have admonished district courts to afford local governments a reasonable opportunity to propose a constitutionally permissible plan and not haphazardly to order injunctive relief. 26 Id. (citing Wise v. Lipscomb, 437 U.S. 535, 540, 98 S.Ct. 2493, 2497, 57 L.Ed.2d 411 (1978)). The extraordinary relief granted in this case, in which the district court not only ordered the election results overturned, but required the county to recreate and refund the eliminated constable office, is reserved only for the most extreme cases. See, e.g., Cook v. Luckett, 735 F.2d 912, 922 (5th Cir.1984). This case was in no way extreme. The court’s order was an abuse of discretion. B. Plaintiffs’ Constitutional Claim While the district court’s analysis of the plaintiffs’ Section 2 Voting Rights Act claim was wanting in many respects, we note, however, that the district court engaged in a careful analysis of the plaintiffs’ claim under the Texas Constitution. Our review of the briefs and record indicates that the district court properly resolved this claim on the merits. We affirm this portion of the judgment. III. CONCLUSION This is the rare case in which the district court erroneously refused to consider probative evidence and just as erroneously relied on non-probative evidence to support its vote dilution finding. To uphold a finding of vote dilution without any supporting evidence, and with much evidence that indicates (a) sustained Hispanic electoral success in Precinct Two; (b) significant Anglo crossover voting for Hispanic candidates; and (c) nearly proportional Hispanic representation in the justice precinct posts, ignores modern-day reality. The court’s finding and resulting judgment cannot stand. *871 For the reasons discussed above, the judgment of the district court is REVERSED IN PART and AFFIRMED IN PART. In light of ■ our conclusions, we RENDER JUDGMENT in favor of Bexar County and VACATE the district court’s injunctive order. Footnotes [1] The Section 5 claim was rejected by a three-judge panel in April 2003 and is no longer at issue. [2] Setting aside an election is a drastic remedy. See Bell v. Southwell, 376 F.2d 659, 662 (5th Cir.1967); and Cook v. Luckett, 735 F.2d 912, 921-22 (5th Cir.1984). Such a remedy should only be imposed where timely pre-election relief is either denied or precluded. See Toney v. White, 488 F.2d 310, 313-315. (5th Cir.1973) (en banc); and Saxon v. Fielding, 614 F.2d 78, 79-80 (5th Cir.1980). [3] As the district court recognized, .the one-person, one-vote requirement of Baker v. Carr, 369 U.S. 186, 82 S.Ct. 691, 7 L.Ed.2d 663 (1962), does not apply to judicial districts like Justice Precincts. Wells v. Edwards, 347 F.Supp. 453, 454 (M.D.La.1972) aff'd, 409 U.S. 1095, 93 S.Ct. 904, 34 L.Ed.2d 679 (1973). However, the district court, at various points in its opinion, expressed concern regarding the application of the Gingles threshold test to single-member districts that are not required to comply with the one-.person, one-vote requirement. Since Section 2 includes judicial selections, Chisom v. Roemer, 501 U.S. 380, 404, 111 S.Ct. 2354, 2369, 115 L.Ed.2d 348 (1991), we are at a loss as to what other- standard than Gingles might apply. [4] All references in this opinion to Justice Precinct Two, unless otherwise noted, refer to the newly redistricted Justice Precinct Two. [5] This court has repeatedly endorsed the analysis of exogenous elections in Section 2 vote dilution cases. See Rangel v. Morales, 8 F.3d 242, 247 (5th Cir.1993); NAACP v. Fordice, 252 F.3d 361, 370 (5th Cir.2001). [6] See Johnson v. Miller, 864 F.Supp. 1354, 1391 (S.D.Ga.1994) (per curiam) (“Statistically speaking, reconstituted election results from precincts within a certain district, actual prior election results from a certain district, and frequency distributions are the primary methods used to estimate the percentages needed to give [minority] voters an equal opportunity to elect a candidate of their choice.”) aff'd Miller v. Johnson, 515 U.S. 900, 115 S.Ct. 2475, 132 L.Ed.2d 762 (1995). [7] The defendant's expert, Dr. John R. Alford, also examined reconstituted election results from nine races from the 2000 general election and eight races from the 1998 general election, as well as two races from the 1998 Democratic primary election and seven races from the 2002 Democratic primary election. See Ex. D-9. [8] In using the term "won,” Dr. Flores apparently means that the candidate obtained more than 50 percent of the votes cast based on his flawed calculation methods. Beyond the simple arithmetic error made by Dr. Flores, the use of a 50 percent threshold may, in some cases, misrepresent the actual percentage necessary to win because it fails to account for the potential presence of a third candidate. As a result, even if Dr. Flores had employed the proper denominator, his approach would not accurately indicate which candidate actually "won,” because while Dr. Flores is correct that a candidate "wins” if he obtain more than 50 percent of the vote in a two-person, head-to-head race, a candidate may win a three-way race with as little as 34 percent of the vote (assuming the other two candidates split the vote evenly). For example, in two of the races analyzed by both experts in the 2002 election — the Governor's race and the race for Court of Criminal Appeals, Place 3 — the winning candidate within new Justice Precinct Two, i.e.. the candidate who got the most votes within the target area, nonetheless garnered less than 50 percent of the total votes cast in that race within the target area. See Ex. D-9. In this sense, Dr. Flores's report not only improperly derives the candidates' vote percentages, but also misrepresents the threshold portion of the vote necessary to win in a given race. [9] Indeed, even the one loss was extremely close, with a difference of only 26 votes out of a total of 79,888 votes cast in that race. See Ex. D-9, table 4. [10] The exact numbers contained in defense exhibit D-13 vary slightly from the numbers in the trial transcript apparently because of the way that state law requires early votes from small precincts to be reported. These differences were discussed at trial and do hot make a significant difference in the percentages discussed above. See ROA vol. 11 at 74-76, Ex. D-13. [11] We note that this is not the first time that this court has found substantial errors in Dr. Flores’s work. See Rollins v. Fort Bend Indep. Sch. Dist., 89 F.3d 1205, 1214-15 (5th Cir.1996) ("FBISD demonstrated inconsistencies in Dr. Flores’s data and showed that some of Dr. Flores’s methodologies made his results inaccurate or unreliable. Dr. Flores manually corrected exhibits while testifying and admitted to other errors FBISD and the district court identified. Dr. Flores's testimony also indicated that his analysis was incomplete and slanted in support of the black plaintiffs .... Dr. Flores ... [was] forced to concede that several of [his] opinions were either suspect or incorrect.”). In Rollins, this court upheld the district court's decision to discredit Dr. Flores’s findings and theories on the basis of numerous errors in his analysis. Id. at 1219. [12] In contrast, the plaintiffs’ expert testified that the 2000 election represented an extremely favorable electoral situation for Republicans. See ROA vol. 11 at 67. Given that the Hispanic candidate of choice—according to both parties’ experts—is the Democratic candidate, regardless of ethnicity, it also makes little sense to use this "special circumstance” to explain the success of Hispanic-favored candidates. See id. at 64 (plaintiffs’ expert agreeing with defense counsel's assertion that "in Precinct 2 general elections, the Hispanic candidate of choice is almost certain to be a Democrat” and that “Hispanic Republicans were not the candidate[s] of Hispanic choice.... Hispanics, when given a choice, voted for the Democrat[.]”). [13] While statistical evidence is not always conclusive in a racial polarization inquiry, where the record supports no other conclusion than that suggested by the statistical evidence, such evidence has substantial probative value. Cf. Clark v. Calhoun County, 88 F.3d 1393, 1397-98 (5th Cir.), reh’g denied, 95 F.3d 1151 (1996) (finding racially polarized voting where the plaintiffs provided statistical evidence showing such polarization and the trial record did not support a contrary finding). [14] The plaintiffs’ expert examined nine precincts in Bexar County with Hispanic voter registration over 90 percent and (because only one precinct consisting of 14 total registered voters had Anglo voter registration over 90 percent) eight precincts with Anglo voter registration over 80 percent. ROA vol. 11 at 22-24. [15] This is perhaps not surprising given that the district court had (erroneously) held that the 2002 elections used in the plaintiffs' homogenous precinct analysis were not reliable because of the presence of "special circumstances.” [16] Dr. Flores's expert report indicates that he examined approximately 200 different electoral districts in the San Antonio metropolitan area. Ex. P-146, ¶ 4. However, plaintiffs' trial exhibits that list the jurisdictions examined only show a total of 154 jurisdictions, 39 of which are City Council Districts that fall outside the 1992-2001 timeframe that the plaintiffs' expert analyzed. Exs. P-14, P-15, ROA vol. 11 at 26. Thus, the data submitted at trial appear to indicate that the plaintiffs’ expert only examined 115 districts within the relevant timeframe. [17] The district court's confusion in this regard is somewhat understandable given that on direct examination, the plaintiffs' expert testified regarding the results of his homogenous precinct analysis immediately before and after presenting the results of his additional analysis regarding these redistricted single member districts. See ROA vol. 11 at 22-24, 53-56. [18] The use of "Spanish-surname” registration is novel and highly problematic. At least one district court has recently noted the problems associated with "Spanish-surname analysis” because of its tendency to misidentify Hispanic persons as non-Hispanic and vice-versa. See United States v. Alamosa County, 306 F.Supp.2d 1016, 1022 (D.Colo.2004) (Krieger, J.). That court held that the expert testimony based on Spanish-surname data, while probative, should be afforded reduced weight, and noted that self-identification data provides a more reliable means of determining ethnicity. Id. Both parties in this case presented Spanish-surname data and neither argues that the district afforded too much weight to these data. However, part of the testimony at trial, as well as some of the discussion in the briefs before this court, focused on the fact that the 2002 race for Constable in Justice Precinct Two took place between two Hispanics named Garza and Wilborn. The fact that Wilborn, a Hispanic with a "non-Hispanic” name, would not have been counted in the Spanish-surname registered voter data presented by either party, gives us pause as to the reliability of such data. We share the concerns raised by the district court in Alamosa County regarding the use of Spanish-surname data, and agree that without a strict showing of its probativeness, Spanish-surname data are disfavored, and census data based upon self-identification provides the proper basis for analyzing Section 2 vote dilution claims in the future. [19] This category includes the regression and homogenous precinct analysis actually conducted by the plaintiffs' expert that analyzed 13 races from the 2002 election cycle. See ROA vol. 11 at 22-24. [20] At trial, thére was some dispute and/or confusion as to the percentage of Hispanic voters in new Justice Precinct Two. The plaintiffs’ expert initially testified that 48.5 percent of the registered voters were Hispanic, but then later testified that 48.2 percent of the registered voters were Hispanic. See ROA vol. 11 at 13 ("Well, currently, [Justice] Precinct 2, the way it was redistricted, has 48.5 percent Hispanic registered voters."); See ROA vol. 11 at 24 ("You compare that to new [Justice] Precinct 2 and the — where the registration is 48.2 percent ..."). The defendant's expert, on the other hand, testified that the Hispanic registration was 49 percent by 2002. See ROA vol. 12 at 52 ("Q: Now, let's go to the 2002 general election. And first, what's happened to the Spanish surname registered *868 voter level? A: By 2002, you’re at 49 percent.”). The district court found that Hispanic voters made up 48.8 percent of registered voters in new Justice Precinct Two. Because the relatively minor differences in these numbers do not substantively alter our analysis, and neither party directly argues that the district court clearly erred in its factual finding, we accept, for the sake of argument, the district court’s calculation. [21] These "three districts” all represent the same state House district at different points in time. Under Dr. Flores's methodology, these districts were considered separate and distinct. [22] Further undercutting Dr. Flores’s conclusion is his admission at trial that roughly between 25 and 33 percent of Anglo voters would cross over to support the Hispanic candidate of choice within new Justice Precinct Two. See ROA vol. 11 at 65. In Gingles, the Supreme Court made clear that crossover voting by the majority racial group is relevant to the racial polarization inquiry. See Gingles, 478 U.S. at 56, 106 S.Ct. at 2769. In applying the Gingles factors in the context of a racial gerrymandering case, the Supreme Court more recently declined to overturn a district court’s determination that insufficient evidence of racial polarization existed where Anglos crossed over to vote for the minority candidate of choice at rates ranging between 22 and 38 percent. See Abrams v. Johnson, 521 U.S. 74, 92-93, 117 S.Ct. 1925, 1936-37, 138 L.Ed.2d 285 (1997). In the present case, the district court did not even discuss Dr. Flores’s acknowledgment of a relatively substantial portion of Anglo voters’ support for Hispanic candidates. See ROA vol. 11 at 65. [23] Because we hold that the district court erred in determining that 50 percent was an appropriate threshold below which the third Gingles factor was satisfied, we need not reach the question whether the district court was correct in holding that the post-trial submission of Bexar County — indicating that the Hispanic registration in new Justice Precinct Two exceeded 50 percent — "proves nothing.” However, we do note that even had the district court’s determination regarding the threshold percentage been correct, such evidence would strongly indicate that no relief was warranted, given that the primary factor relied upon by the district court in finding liability — Hispanic voter registration below 50 percent — ceased to exist. Cf. Westwego Citizens for Better Gov’t v. City of Westwego, 906 F.2d 1042, 1045 (5th Cir.1990) (remanding to the district court for consideration of post-trial election results that related to the evidence presented at trial and the district court's findings). [24] Q: All right. Now we had some areas of disagreement. Is it fair to say that we’re going to disagree as to whether Anglos in Precinct 2 vote as a bloc to usually defeat the Hispanic choice in nonpresidential year general elections? A: To me, it’s not clear. That’s correct. Q: So you’re saying — let me back up just a second. You're saying that it’s not clear whether they vote as a bloc usually to defeat? A: Well, in non-presidential year general election, all we've got is that one — that one election. So usually in that one election, to me, that doesn’t make very much — if we have a history of elections, then I could — I could have a better answer for this. But I really can't answer this. Q: All right. So we may agree that they don’t vote as a bloc usually to defeat the Hispanic choice or we may not disagree on that.. You just don't know? A: To me, this is an area of uncertainty. [25] Plaintiffs' refusal to identify Constable Wil-born as Hispanic simply because he lacks an Hispanic surname borders on the ridiculous. [26] In our view, the district court did not afford Bexar County a reasonable opportunity to fashion an appropriate remedy. In Bexar County’s filings with the district court regarding the relief to be ordered, as noted above, the county pointed out that the district court’s primary rationale for finding liability — the fact that Hispanic voter registration was under 50 percent in Justice Precinct Two — had changed since the time of trial. See R. Doc. 190 (Bexar County's suggestion to the court on the scope of available relief and reconsideration of Section 2 finding highlighting the changed circumstances); R. Doc. 193 (Bexar Court’s suggestion of mootness based on the changed circumstances). As a result, the County requested that the district court either alter its finding on liability or provide the County with information on what criteria the court would use to evaluate a proposed plan so that the county could submit such a plan. See R. Doc. 190 (requesting that the district court either alter its finding on liability or, in the alternative, provide additional guidance to the parties); R. Doc. 202 (Bexar County's comments on the plaintiffs’ September 18 advisory to the court indicating that Bexar County stood ready “to propose a remedial plan that cures the legal infirmities that have been identified by the court,” but requesting "a status conference or other hearing prior to [the district court] granting relief’ and arguing that such a proceeding "would be of great benefit to the court and the parties in determining exactly what relief would address the infirmities identified by the court.”). However, rather than address Bexar County's reasonable concerns, the district court simply granted the bulk of the plaintiffs’ requested relief. R. Doc. 203. Such an approach does not comport with the Supreme Court's and this court’s clear requirements.