with whom Justice Stevens, Justice Ginsburg, and Justice Breyer join, dissenting.
I — I
I agree with the Court that reducing the number of majority-minority districts within a State would not necessarily amount to retrogression barring preclearance under § 5 of the Voting Rights Act of 1965. See ante, at 480-482. The prudential objective of § 5 is hardly betrayed if a State can show that a new districting plan shifts from supermajority districts, in which minorities can elect their candidates of choice by their own voting power, to coalition districts, in which minorities are in fact shown to have a similar opportunity when joined by predictably supportive nonminority voters. Cf. Johnson v. De Grandy, 512 U. S. 997, 1020 (1994) (explaining in the context of §2 that although “society’s racial and ethnic cleavages sometimes necessitate majority-minority districts to ensure equal political and electoral opportunity, that should not obscure the fact that there are communities in which minority citizens are able to form coalitions with voters from other racial and ethnic groups, having no need to be a majority within a single district in order to elect candidates of their choice”).
Before a State shifts from majority-minority to coalition districts, however, the State bears the burden of proving that nonminority voters will reliably vote along with the minority. See, e. g., Reno v. Bossier Parish School Bd., 520 *493U. S. 471, 478 (1997). It must show not merely that minority voters in new districts may have some influence, but that minority voters will have effective influence translatable into probable election results comparable to what they enjoyed under the existing district scheme. And to demonstrate this, a State must do more than produce reports of minority voting age percentages; it must show that the probable voting behavior of nonminority voters will make coalitions with minorities a real prospect. See, e.g., Pildes, Is Voting-Rights Law Now at War With Itself? Social Science and Voting Rights in the 2000s, 80 N. C. L. Rev. 1517, 1539 (2002). If the State’s evidence fails to convince a factfinder that high racial polarization in voting is unlikely, or that high white crossover voting is likely, or that other political and demographic facts point to probable minority effectiveness, a reduction in supermajority districts must be treated as potentially and fatally retrogressive, the burden of persuasion always being on the State.
The District Court majority perfectly well understood all this and committed no error. Error enters this case here in this Court, whose majority unmoors §5 from any practical and administrable conception of minority influence that would rule out retrogression in a transition from majority-minority districts, and mistakes the significance of the evidence supporting the District Court’s decision.
II
The Court goes beyond recognizing the possibility of coalition districts as nonretrogressive alternatives to those with majorities of minority voters when it redefines effective voting power in § 5 analysis without the anchoring reference to electing a candidate of choice. It does this by alternatively suggesting that a potentially retrogressive redistricting plan could satisfy § 5 if a sufficient number of so-called “influence districts,” in addition to “coalitio[n] districts,” were created, ante, at 483,484, or if the new plan provided minority groups *494with an opportunity to elect a particularly powerful candidate, ante, at 483-484. On either alternative, the §5 requirement that voting changes be nonretrogressive is substantially diminished and left practically unadministrable.
A
The Court holds that a State can carry its burden to show a nonretrogressive degree of minority “influence” by demonstrating that “ ‘candidates elected without decisive minority support would be willing to take the minority’s interests into account.’” Ante, at 482 (quoting Thornburg v. Gingles, 478 U. S. 30, 100 (1986) (O’Connor, J., concurring in judgment)). But this cannot be right.
The history of § 5 demonstrates that it addresses changes in state law intended to perpetuate the exclusion of minority voters from the exercise of political power. When this Court held that a State must show that any change in voting procedure is free of retrogression it meant that changes must not leave minority voters with less chance to be effective in electing preferred candidates than they were before the change. “[T]he purpose of § 5 has always been to insure that no voting-procedure changes would be made that would lead to a retrogression in the position of racial minorities with respect to their effective exercise of the electoral franchise.” Beer v. United States, 426 U. S. 130, 141 (1976); see, e. g., id., at 140-141 (“Section 5 was intended ‘to insure that [the gains thus far achieved in minority political participation] shall not be destroyed through new [discriminatory] procedures and techniques’” (quoting S. Rep. No. 94-295, p. 19 (1975))). In addressing the burden to show no retrogression, therefore, “influence” must mean an opportunity to exercise power effectively.
The Court, however, says that influence may be adequate to avoid retrogression from majority-minority districts when it consists not of decisive minority voting power but of sentiment on the part of politicians: influence may be sufficient *495when it reflects a willingness on the part of politicians to consider the interests of minority voters, even when they do not need the minority votes to be elected. The Court holds, in other words, that there would be no retrogression when the power of a voting majority of minority voters is eliminated, so long as elected politicians can be expected to give some consideration to minority interests.
The power to elect a candidate of choice has been forgotten; voting power has been forgotten. It is very hard to see anything left of the standard of nonretrogression, and it is no surprise that the Court’s cited precedential support for this reconception, see ante, at 482, consists of a footnote from a dissenting opinion in Shaw v. Hunt, 517 U. S. 899 (1996), and footnote dictum in a case from the Western District of Louisiana.
Indeed, to see the trouble ahead, one need only ask how on the Court’s new understanding, state legislators or federal preclearance reviewers under § 5 are supposed to identify or measure the degree of influence necessary to avoid the retrogression the Court nominally retains as the §5 touchstone. Is the test purely ad hominem, looking merely to the apparent sentiments of incumbents who might run in the new districts? Would it be enough for a State to show that an incumbent had previously promised to consider minority interests before voting on legislative measures? Whatever one looks to, however, how does one put a value on influence that falls short of decisive influence through coalition? Non-decisive influence is worth less than majority-minority control, but how much less? Would two influence districts offset the loss of one majority-minority district? Would it take three? Or four? The Court gives no guidance for measuring influence that falls short of the voting strength of a coalition member, let alone a majority of minority voters. Nor do I see how the Court could possibly give any such guidance. The Court’s “influence” is simply not functional in the political and judicial worlds.
*496B
Identical problems of comparability and administrability count at least as much against the Court’s further gloss on nonretrogression, in its novel holding that a State may trade off minority voters’ ability to elect a candidate of their choice against their ability to exert some undefined degree of influence over a candidate likely to occupy a position of official legislative power. See ante, at 483-484. The Court implies that one majority-minority district in which minority voters could elect a legislative leader could replace a larger number of majority-minority districts with ordinary candidates, without retrogression of overall minority voting strength. Under this approach to § 5, a State may value minority votes in a district in which a potential committee chairman might be elected differently from minority votes in a district with ordinary candidates.
It is impossible to believe that Congress could ever have imagined § 5 preclearance actually turning on any such distinctions. In any event, if the Court is going to allow a State to weigh minority votes by the ambitiousness of candidates the votes might be cast for, it is hard to see any stopping point. I suppose the Court would not go so far as to give extra points to an incumbent with the charisma to attract a legislative following, but would it value all committee chairmen equally? (The committee chairmen certainly would not.) And what about a legislator with a network of influence that has made him a proven dealmaker? Thus, again, the problem of measurement: is a shift from 10 majority-minority districts to 8 offset by a good chance that 1 of the 8 may elect a new Speaker of the House?
I do not fault the Court for having no answers to these questions, for there are no answers of any use under §5. The fault is more fundamental, and the very fact that the Court’s interpretation of nonretrogression under §5 invites unanswerable questions points to the error of a § 5 preclearance regime that defies reviewable administration. We are *497left with little hope of determining practically whether a dis-tricting shift to one party’s overall political advantage can be expected to offset a loss of majority-minority voting power in particular districts; there will simply be greater opportunity to reduce minority voting strength in the guise of obtaining party advantage.
One is left to ask who will suffer most from the Court’s new and unquantifiable standard. If it should turn out that an actual, serious burden of persuasion remains on the States, States that rely on the new theory of influence should be guaranteed losers: nonretrogression cannot be demonstrated by districts with minority influence too amorphous for objective comparison. But that outcome is unlikely, and if in subsequent cases the Court allows the State’s burden to be satisfied on the pretense that unquantifiable influence can be equated with majority-minority power, §5 will simply drop out as a safeguard against the “unremitting and ingenious defiance of the Constitution” that required the procedure of preelearance in the first place. South Carolina v. Katzenbach, 383 U. S. 301, 309 (1966).
IIII
The District Court never reached the question the Court addresses, of what kind of influence districts (coalition or not) might demonstrate that a decrease in majority-minority districts was not retrogressive. It did not reach this question because it found that the State had not satisfied its burden of persuasion on an issue that should be crucial on any ad-ministrable theory:1 the State had not shown the possibility *498of actual coalitions in the affected districts that would allow any retreat from majority-minority districts without a retrogressive effect. This central evidentiary finding is invulnerable under the correct standard of review.
This Court’s review of the District Court’s factual findings is for clear error. See, e. g., Miller v. Johnson, 515 U. S. 900, 917 (1995); Pleasant Grove v. United States, 479 U. S. 462, 469 (1987); McCain v. Lybrand, 465 U. S. 236, 258 (1984); City of Lockhart v. United States, 460 U. S. 125, 136 (1983). We have no business disturbing the District Court’s ruling “simply because we would have decided the case differently,” but only if based “on the entire evidence, [we are] left with the definite and firm conviction that a mistake has been committed.” Easley v. Cromartie, 532 U. S. 234, 242 (2001) (internal quotation marks omitted). It is not, then, up to us to “decide whether Georgia’s State Senate redistricting plan is retrogressive as compared to its previous, benchmark dis-tricting plan.” Ante, at 466. Our sole responsibility is to see whether the District Court committed clear error in refusing to preclear the plan. It did not.
A
The District Court began with the acknowledgment (to which we would all assent) that the simple fact of a decrease in black voting age population (BVAP) in some districts is not alone dispositive about whether a proposed plan is retrogressive:
*499“ ‘Unpacking’ African American districts may have positive or negative consequences for the statewide electoral strength of African American voters. To the extent that voting patterns suggest that minority voters are in a better position to join forces with other segments of the population to elect minority preferred candidates, a decrease in a district’s BVAP may have little or no effect on minority voting strength.” 195 F. Supp. 2d 25, 76 (DC 2002).
See id., at 78 (“[T]he Voting Rights Act allows states to adopt plans that move minorities out of districts in which they formerly constituted a majority of the voting population, provided that racial divisions have healed to the point that numerical reductions will not necessarily translate into reductions in electoral power”); id., at 84 (“[T]he mere fact that BVAP decreases in certain districts is not enough to deny preclearance to a plan under Section 5”).2
The District Court recognized that the key to understanding the impact of drops in a district’s BVAP on the minority group’s “effective exercise of the electoral franchise,” Beer, 425 U. S., at 141, is the level of racial polarization. If racial elements consistently vote in separate blocs, decreasing the proportion of black voters will generally reduce the chance that the minority group’s favored candidate will be elected; whereas in districts with low racial bloc voting or significant white crossover voting, a decrease in the black proportion may have no effect at all on the minority’s opportunity to elect their candidate of choice. See, e. g., 195 F. Supp. 2d, at 84 (“[R]acial polarization is critically important because its presence or absence in the Senate Districts challenged by the United States goes a long way to determining whether *500or not the decreases in BVAP and African American voter registration in those districts are likely to produce retrogressive effects”).
This indisputable recognition, that context determines the effect of decreasing minority numbers for purposes of the § 5 enquiry, points to the nub of this case, and the District Court’s decision boils down to a judgment about what the evidence showed about that context. The District Court found that the United States had offered evidence of racial polarization in the contested districts,3 id., at 86, and it found that Georgia had failed to present anything relevant on that issue. Georgia, the District Court said, had “provided the court with no competent, comprehensive information regarding white crossover voting or levels of polarization in individual districts across the State.” Id., at 88. In particular, the District Court found it “impossible to extrapolate” anything about the level of racial polarization from the statistical submissions of Georgia’s lone expert witness. Id., at 85. And the panel majority took note that Georgia’s expert “admitted on cross-examination” that his evidence simply did not address racial polarization: “the whole point of my analysis,” the expert stated, “is not to look at polarization per se. The question is not whether or not blacks and whites in general vote for different candidates.” Ibid. (internal quotation marks omitted).
Accordingly, the District Court explained that Georgia’s expert:
*501“made no attempt to address the central issue before the court: whether the State’s proposal is retrogressive. He failed even to identify the decreases in BVAP that would occur under the proposed plan, and certainly did not identify corresponding reductions in the electability of African American candidates of choice. The paucity of information in [the expert’s] report thus leaves us unable to use his analysis to assess the expected change in African American voting strength statewide that will be brought by the proposed Senate plan.” Id., at 81.
B
How is it, then, that the majority of this Court speaks of “Georgia’s evidence that the Senate plan as a whole is not retrogressive,” against which “the United States did not introduce any evidence [in] rebut[tal],” ante, at 487? The answer is that the Court is not engaging in review for clear error. Instead, it is reweighing evidence de novo, discovering what it thinks the District Court overlooked, and drawing evidentiary conclusions the District Court supposedly did not see. The Court is mistaken on all points.
1
Implicitly recognizing that evidence of voting behavior by majority voters is crucial to any showing of nonretrogression when minority numbers drop under a proposed plan, the Court tries to find evidence to fill the record’s gap. It says, for example, that “Georgia introduced evidence showing that approximately one-third of white voters would support a black candidate in [the contested] districts.” Ante, at 486. In support of this claim, however, the majority focuses on testimony offered by Georgia’s expert relating to crossover voting in the pre-existing rather than proposed districts. 195 F. Supp. 2d, at 66. The District Court specifically noted that the expert did not calculate crossover voting under the proposed plan. Id., at 65, n. 31 (“The court also emphasizes *502that Epstein did not attempt to rely on the table’s calculations to demonstrate voting patterns in the districts, and calculated crossover in the existing, and not the proposed, Senate districts”). Indeed, in relying on this evidence the majority attributes a significance to it that Georgia’s own expert disclaimed, as the District Court pointed out. See id., at 85 (“[I]t is impossible to extrapolate these voting patterns from Epstein’s database. As Epstein admitted on cross-examination: the whole point of my analysis is not to look at polarization per se. The question is not whether or not blacks and whites in general vote for different candidates” (internal quotation marks omitted)).
2
In another effort to revise the record, the Court faults the District Court, alleging that it “focused too narrowly on proposed Senate Districts 2, 12, and 26.” Ante, at 485. In fact, however, it is Georgia that asked the District Court to consider only the contested districts, and the District Court explicitly refused to limit its review in any such fashion: “we reject the State’s argument that this court’s review is limited only to those districts challenged by the United States, and should not encompass the redistricting plans in their entirety. .. . [T]he court’s review necessarily extends to the entire proposed plan.” 195 F. Supp. 2d, at 78. The District Court explained that it “is vested with the final authority to approve or disapprove the proposed change as a whole.” Ibid. “The question before us is whether the proposed Senate plan as a whole, has the ‘purpose or effect of denying or abridging the right to vote on account of race or color.’” Id., at 103 (Oberdorfer, J., concurring in part and dissenting in part) (quoting 42 U. S. C. § 1973c). Though the majority asserts that “[t]he District Court ignored the evidence of numerous other districts showing an increase in black voting age population,” ante, at 486, the District Court, in fact, specifically considered the parties’ dispute over the statewide *503impact of the change in black voting age population. See, e. g., 195 F. Supp. 2d, at 93 (“The number of Senate Districts with majorities of BVAP would, according to Georgia’s calculations, increase from twelve to thirteen; according to the Attorney General’s interpretation of the census data, the number would decrease from twelve to eleven”).
3
In a further try to improve the record, the Court focuses on the testimony of certain lay witnesses, politicians presented by the State to support its claim that the Senate plan is not retrogressive. Georgia, indeed, relied heavily on the near unanimity of minority legislators’ support for the plan. But the District Court did not overlook this evidence; it simply found it inadequate to carry the State’s burden of showing nonretrogression. The District Court majority explained that the “legislators’ support is, in the end, far more probative of a lack of retrogressive purpose than of an absence of retrogressive effect.” Id., at 89 (emphasis in original). As against the politicians’ testimony, the District Court had contrary “credible,” id., at 88, evidence of retrogressive effect. This evidence was the testimony of the expert witness presented by the United States, which “suggests the existence of highly racially polarized voting in the proposed districts,” ibid., evidence of retrogressive effect to which Georgia offered “no competent” response, ibid. The District Court was clearly within bounds in finding that (1) Georgia’s proposed plan decreased BVAP in the relevant districts, (2) the United States offered evidence of significant facial polarization in those districts, and (3) Georgia offered no adequate response to this evidence.
The reasonableness of the District Court’s treatment of the evidence is underscored in its concluding reflection that it was possible Georgia could have shown the plan to be non-retrogressive, but the evidence the State had actually offered simply failed to do that. “There are, without doubt, *504numerous other ways, given the limited evidence of racially polarized voting in State Senate and local elections, that Georgia could have met its burden of proof in this case. Yet, the court is limited to reviewing the evidence presented by the parties, and is compelled to hold that the State has not met its burden.” Id., at 94. “[T]he lack of positive racial polarization data was the gap at the center of the State’s case [and] the evidence presented by [the] estimable [legislators] does not come close to filling that void.” Id., at 100.
As must be plain, in overturning the District Court’s thoughtful consideration of the evidence before it, the majority of this Court is simply rejecting the District Court’s evi-dentiary finding in favor of its own. It is reweighing testimony and making judgments about the competence, interest, and character of witnesses. The Court is not conducting clear error review.
4
Next, the Court attempts to fill the holes in the State’s evidence on retrogression by drawing inferences favorable to the State from undisputed statistics. See ante, at 487-489. This exercise comes no closer to demonstrating clear error than the others considered so far.
In the first place, the District Court has already explained the futility of the Court’s effort. Knowing whether the number of majority BVAP districts increases, decreases, or stays the same under a proposed plan does not alone allow any firm conclusion that minorities will have a better, or worse, or unvarying opportunity to elect their candidates of choice. Any such inference must depend not only on trends in BVAP levels, but on evidence of likely voter turnout among minority and majority groups, patterns of racial bloc voting, likelihood of white crossover voting, and so on.4 In*505deed, the core holding of the Court today, with which I agree, that nonretrogression does not necessarily require maintenance of existing supermajority minority districts, turns on this very point; comparing the number of majority-minority districts under existing and proposed plans does not alone reliably indicate whether the new plan is retrogressive.
Lack of contextual evidence is not, however, the only flaw in the Court’s numerical arguments. Thus, in its first example, ante, at 487, the Court points out that under the proposed plan the number of districts with majority BVAP increases by one over the existing plan,5 but the Court does not mention that the number of districts with BVAP levels over 55% decreases by four. See Record, Doc. No. 148, Pl. Exhs. ID, 2C. Similarly, the Court points to an increase of two in districts with BVAP in the 30% to 50% range, along with a further increase of two in the 25% to 30% range. Ante, at 487. It fails to mention, however, that Georgia’s own expert argued that 44.3% was the critical threshold for BVAP levels, 195 F. Supp. 2d, at 107, and the data on which the Court relies shows the number of districts with BVAP over 40% actually decreasing by one, see Record, Doc. No. 148, Pl. Exhs. ID, 2C. My point is not that these figures conclusively demonstrate retrogression; I mean to say only that percentages tell us nothing in isolation, and that without contextual evidence the raw facts about population levels fail to get close to indicating that the State carried its burden to show no retrogression. They do not come close to showing clear error.
*5065
Nor could error, clear or otherwise, be shown by the Court’s comparison of the proposed plan with the description of the State and its districts provided by the 1990 census. Ante, at 487-489. The 1990 census is irrelevant. We have the 2000 census, and precedent confirms in no uncertain terms that the issue for § 5 purposes is not whether Georgia’s proposed plan would have had a retrogressive effect 13 years ago: the question is whether the proposed plan would be retrogressive now. See, e. g., Reno v. Bossier Parish School Bd., 528 U. S. 320, 334 (2000) (Under §5 “the baseline is the status quo that is proposed to be changed”); Holder v. Hall, 512 U. S. 874, 883 (1994) (plurality opinion) (Under § 5, “[t]he baseline for comparison is present by definition; it is the existing status”); City of Lockhart v. United States, 460 U. S., at 132 (“The proper comparison is between the new system and the system actually in effect”); Cf. 28 CFR § 51.54(b)(2) (2002) (when determining if a change is retrogressive under § 5 “[t]he Attorney General will make the comparison based on the conditions existing at the time of the submission”). The Court’s assumption that a proper § 5 analysis may proceed on the basis of obsolete data from a superseded census is thus as puzzling as it is unprecedented. It is also an invitation to perverse results, for if a State could carry its burden under § 5 merely by showing no retrogression from the state of affairs 13 years ago, it could demand preelearance for a plan flatly diminishing minority voting strength under §5.6
*5076
The Court’s final effort to demonstrate that Georgia’s plan is nonretrogressive focuses on statistics about Georgia Democrats. Ante, at 489. The Court explains that almost all the districts in the proposed plan with a BVAP above 20% have a likely overall Democratic performance above 50%, and from this the Court concludes that “[t]hese statistics make it more likely as a matter of fact that black voters will constitute an effective voting bloc.” Ibid. But this is not so. The degree to which the statistics could support any judgment about the effect of black voting in State Senate elections is doubtful, and even on the Court’s assumptions the statistics show no clear error by the District Court.
As for doubt about what the numbers have to do with State Senate elections, it is enough to know that the majority’s figures are taken from a table describing Democratic voting in statewide, not local, elections. The Court offers no basis for assuming that voting for Democratic candidates in statewide elections correlates with voting behavior in local elections,7 and in fact, the record points to different, not identical, voting patterns. The District Court specifically noted that the United States’s expert testified that “African American candidates consistently received less crossover voting in local election[s] than in statewide elections,” 195 F. Supp. 2d, at 71, and the court concluded that there is “compelling evidence that racial voting patterns in State Senate races can be expected to differ from racial voting patterns in statewide races,” id., at 85-86.
*508But even if we assume the data on Democratic voting statewide can tell us something useful about Democratic voting in State Senate districts, the Court’s argument does not hold up. It proceeds from the faulty premise that even with a low BVAP, if enough of the district is Democratic, the minority Democrats will necessarily have an effect on which candidates are elected. But if the proportion of nonminority Democrats is high enough, the minority group may well have no impact whatever on which Democratic candidate is selected to run and ultimately elected. In districts, say, with 20% minority voters (all of them Democrats) and 51% nonmi-nority Democrats, the Democratic candidate has no obvious need to take the interests of the minority group into account; if everybody votes (or the proportion of stay-at-homes is constant throughout the electorate) the Democrat can win the general election without minority support. Even in a situation where a Democratic candidate needs a substantial fraction of minority voters to win (say the population is 25% minority and 30% nonminority Democrats), the Democratic candidate may still be able to ignore minority interests if there is such ideological polarization as between the major parties that the Republican candidate is entirely unresponsive to minority interests. In that situation, a minority bloc would presumably still prefer the Democrat, who would not need to adjust any political positions to get the minority vote.
All of this reasoning, of course, carries a whiff of the lamp. I do not know how Georgia’s voters will actually behave if the percentage of something is x, or maybe y, any more than the Court does. We are arguing about numerical abstractions, and my sole point is that the Court’s abstract arguments do not hold up. Much less do they prove the District Court wrong.
IV
Section 5, after all, was not enacted to address abstractions. It was enacted “to shift the advantage of time and *509inertia from the perpetrators of the evil to its victim,” Beer, 425 U. S., at 140 (internal quotation marks omitted) (quoting H. R. Rep. No. 94-196, pp. 57-58 (1970)), and the State of Georgia was made subject to the requirement of preclearance because Congress “had reason to suppose” it might “try ... to evade the remedies for voting discrimination” and thus justifies §5’s “uncommon exercise of congressional power.” South Carolina v. Katzenbach, 383 U. S., at 334-335. Section 5 can only be addressed, and the burden to prove no retrogression can only be carried, with evidence of how particular populations of voters will probably act in the circumstances in which they live. The State has the burden to convince on the basis of such evidence. The District Court considered such evidence: it received testimony, decided what it was worth, and concluded as the trier of fact that the State had failed to carry its burden. There was no error, and I respectfully dissent.
The District Court correctly recognized that the State bears the burden of proof in establishing that its proposed redistricting plan satisfied the standards of § 5. See, e. g., 195 F. Supp. 2d 25, 86 (DC 2002) (“We look to the State to explain why retrogression is not present”); see also Reno v. Bossier Parish School Bd., 520 U. S. 471, 478 (1997) (covered jurisdiction “bears the burden of proving that the change does not have the purpose and will not have the effect of denying or abridging the right to vote on *498account of race or color” (internal quotation marks omitted)); id., at 480 (Section 5 “imposes upon a covered jurisdiction the difficult burden of proving the absence of discriminatory purpose and effect”); Reno v. Bossier Parish School Bd., 528 U. S. 320, 332 (2000) (“In the specific context of § 5 . . . the covered jurisdiction has the burden of persuasion”); cf. Beer v. United States, 425 U. S. 130, 140 (1976) (Congress in passing §5 sought to “freez[e] election procedures in the covered areas unless the changes can be shown to be nondiscriminatory” (internal quotation marks omitted)).
Indeed, the other plans approved by the District Court, Georgia’s State House plan, 195 F. Supp. 2d, at 95, congressional plan, ibid., and the interim plan approved for the State Senate, 204 F. Supp. 2d 4, 7 (DC 2002), all included decreases in BVAP in particular districts.
The majority cites the District Court’s comment that ‘“the United States’ evidence was extremely limited in scope — focusing only on three contested districts in the State Senate plan.’ ” Ante, at 474 (quoting 195 F. Supp. 2d, at 37). The District Court correctly did not require the United States to prove that the plan was retrogressive. As the District Court explained: “[u]ltimately, the burden of proof in this matter lies with the State. We look to the State to explain why retrogression is not present, and to prove the absence of racially polarized voting that might diminish African American voting strength in light of several districts’ decreased BVAPs.” Id., at 86.
The fact that the Court premises its analysis on BVAP alone is ironic given that the Court, incorrectly, chastises the District Court for committing the very error the Court now engages in, ‘Tailing] to consider all the relevant factors.” Ante, at 485.
Though the Court does not acknowledge it in its discussion of why “Georgia likely met its burden,” ante, at 487, even this claim was disputed. As the District Court explained: “[tjhe number of Senate Districts with majorities of BVAP would, according to Georgia’s calculations, increase from twelve to thirteen; according to the Attorney General’s interpretation of the census data, the number would decrease from twelve to eleven.” 195 F. Supp. 2d, at 93.
For example, if a covered jurisdiction had two majority-minority districts in 1990, but rapidly changing demography had produced two more during the ensuing decade, a new redistricting plan, setting the number of majority-minority districts at three would conclusively rule out retrogression on the Court’s calculus. This would be the case even when voting behavior showed that nothing short of four majority-minority districts would preserve the status quo as of 2000.
Even if the majority wanted to rely on these figures to make a claim about Democratic voting in statewide elections, the predictors’ significance is utterly unclear. The majority pulls its figures from an exhibit titled, “Political Data Report,” and a column labeled, “%OVER DEMVOTES,” Pl. Exh. 2D. See ante, at 489. The document provides no information regarding whether the numbers in the column reflect an average of past performance, a prediction for future performance, or something else altogether.