Justice Breyer
delivered the opinion of the Court.
In this appeal, we review a three-judge District Court’s determination that North Carolina’s Legislature used race as the “predominant factor” in drawing its 12th Congressional District’s 1997 boundaries. The court’s findings, in our view, are clearly erroneous. We therefore reverse its conclusion that the State violated the Equal Protection Clause. U. S. Const., Arndt. 14, § 1.
I
This “racial districting” litigation is before us for the fourth time. Our first two holdings addressed North Carolina’s former Congressional District 12, one of two North Carolina congressional districts drawn in 1992 that contained a majority of African-American voters. See Shaw v. Reno, 509 U.S. 630 (1993) (Shaw I); Shaw v. Hunt, 517 U.S. 899 (1996) (Shaw II).
A
In Shaw I, the Court considered whether plaintiffs’ factual allegation — that the legislature had drawn the former district’s boundaries for race-based reasons — if true, could underlie a legal holding that the legislature had violated the Equal Protection Clause. The Court held that it could. It wrote that a violation may exist where the legislature’s boundary drawing, though “race neutral on its face,” none[238] theless can be understood only as an effort to “separate voters into different districts on the basis of race,” and where the “separation lacks sufficient justification.” 509 U. S., at 649.
In Shaw II, the Court reversed a subsequent three-judge District Court’s holding that the boundary-drawing law in question did not violate the Constitution. This Court found that the district’s “unconventional,” snakelike shape, the way in which its boundaries split towns and counties, its predominately African-American racial makeup, and its history, together demonstrated a deliberate effort to create a “majority-black” district in which race “could not be compromised,” not simply a district designed to “protec[t] Democratic incumbents.” 517 U. S., at 902-90B, 905-907. And the Court concluded that the legislature’s use of racial criteria was not justified. Id., at 909-918.
B
Our third holding focused on a new District 12, the boundaries of which the legislature had redrawn in 1997. Hunt v. Cromartie, 526 U.S. 541 (1999). A three-judge District Court, with one judge dissenting, had granted summary judgment in favor of those challenging the district’s boundaries. The court found that the legislature again had “used criteria . . . that are facially race driven,” in violation of the Equal Protection Clause. App. to Juris. Statement in No. 99-1864, p. 262a (hereinafter App. to Juris. Statement). It based this conclusion upon “uncontroverted material facts” showing that the boundaries created an unusually shaped district, split counties and cities, and in particular placed almost all heavily Democratic-registered, predominantly African-American voting preeinets, inside the district while locating some heavily Democratic-registered, predominantly white precincts, outside the district. This latter circumstance, said the court, showed that the legislature was trying to maximize new District 12’s African-[239] American voting strength, not the district’s Democratic voting strength. Ibid.
This Court reversed. We agreed with the District Court that the new district’s shape, the way in which it split towns and counties, and its heavily African-American voting population all helped the plaintiffs’ case. 526 U.S., at 547-549. But neither that evidence by itself, nor when coupled with the evidence of Democratic registration, was sufficient to show, on summary judgment, the unconstitutional race-based objective that plaintiffs claimed. That is because there was a genuine issue of material fact as to whether the evidence also was consistent with a constitutional political objective, namely, the creation of a safe Democratic seat. Id., at 549-551.
We pointed to the affidavit of an expert witness for defendants, Dr. David W. Peterson. Dr. Peterson offered to show that, because North Carolina’s African-American voters are overwhelmingly Democratic voters, one cannot easily distinguish a legislative effort to create a majority-African-Ameriean district from a legislative effort to create a safely Democratic district. Id., at 550. And he also provided data showing that registration did not indicate how voters would actually vote. Id., at 550-551. We agreed that data showing how voters actually behave, not data showing only how those voters are registered, could affect the outcome of this litigation. Ibid. We concluded that the case was “not suited for summary disposition” and we reversed the District Court. Id., at 554.
C
On remand, the parties undertook additional discovery. The three-judge District Court held a 3-day trial. And the court again held (over a dissent) that the legislature had unconstitutionally drawn District 12’s new 1997 boundaries. It found that the legislature had tried “(1) [to] eur[e] the [previous district’s] constitutional defects” while also “(2) drawing the plan to maintain the existing partisan bal-[240] anee in the State’s congressional delegation.” Cromartie v. Hunt, 133 F. Supp. 2d 407, 413 (EDNC 2000). It added that to "achieve the second goal,” the legislature “drew the new plan (1) to avoid placing two incumbents in the same district and (2) to preserve the partisan core of the existing districts.” Ibid. The court concluded that the "plan as enacted largely reflects these directives.” Ibid. But the court also found “as a matter of fact that the General Assembly ... used criteria... that are facially race driven” without any compelling justification for doing so. Id., at 420.
The court based its latter, constitutionally critical, conclusion in part upon the district’s snakelike shape, the way in which it split cities and towns, and its heavily African-American (47%) voting population, id., at 413-415 — all matters that this Court had considered when it found summary judgment inappropriate, Cromartie, 526 U.S., at 544. The court also based this conclusion upon a specific finding — absent when we previously considered this litigation — that the legislature had drawn the boundaries in order “to collect precincts with high racial identification rather than political identification.” 133 F. Supp. 2d, at 420 (emphasis added).
This last-mentioned finding rested in turn upon five subsidiary determinations:
(1) that “the legislators excluded many heavily-Democratie precincts from District 12, even when those precincts immediately border the Twelfth and would have established a far more compact district,” id., at 419; see also id., at 421 ("more heavily Democratic precincts . . . were bypassed... in favor of precincts with a higher African-American population”);
(2) that “[additionally, Plaintiffs’ expert, Dr. Weber, showed time and again how raee trumped party affiliation in the construction of the 12th District and how political explanations utterly failed to explain the composition of the district,” id., at 419;
[241] (3) that Dr. Peterson’s testimony was “ ‘unreliable’ and not relevant,” id., at 420 (citing testimony of Dr. Weber);
(4) that a legislative redistricting leader, Senator Roy Cooper, had alluded at the time of redistrieting “to a need for ‘racial and partisan’ balance,” ibid.; and
(5) that the Senate’s redistricting coordinator, Gerry Cohen, had sent Senator Cooper an e-mail reporting that Cooper had “moved Greensboro Black community into the 12th, and now need[ed] to take [about] 60,000 out of the 12th,” App. 369; 133 P. Supp. 2d, at 420.
The State and intervenors filed a notice of appeal. 28 U. S. C. § 1253. We noted probable jurisdiction. 530 U. S. 1260 (2000). And we now reverse.
II
The issue in this case is evidentiary. We must determine whether there is adequate support for the District Court’s key findings, particularly the ultimate finding that the legislature’s motive was predominantly racial, not political. In making this determination, we are aware that, under Shaw I and later cases, the burden of proof on the plaintiffs (who attack the district) is a “demanding one.” Miller v. Johnson, 515 U.S. 900, 928 (1995) (O’Connor, J., concurring). The Court has specified that those who claim that a legislature has improperly used race as a criterion, in order, for example, to create a majority-minority district, must show at a minimum that the “legislature subordinated traditional race-neutral districting principles ... to racial considerations.” Id., at 916 (majority opinion). Race must not simply have been “a motivation for the drawing of a majority-minority district,” Bush v. Vera, 517 U.S. 952, 959 (1996) (O’Connor, J., principal opinion) (emphasis in original), but “the ‘predominant factor’ motivating the legislature’s districting decision,” Cromartie, swpfa, at 547 (quoting Miller, supra, at 916) (emphasis added). Plaintiffs [242] must show that a facially neutral law “'is “unexplainable on grounds other than race.”'” Cromartie, supra, at 546 (quoting Shaw I, 509 U. S., at 644, in turn quoting Arlington Heights v. Metropolitan Housing Development Corp., 429 U.S. 252, 266 (1977)).
The Court also has made clear that the underlying dis-tricting decision is one that ordinarily falls within a legislature’s sphere of competence. Miller, 515 U.S., at 915. Hence, the legislature “must have discretion to exercise the political judgment necessary to balance competing interests,” ibid., and courts must “exercise extraordinary caution in adjudicating claims that a State has drawn district lines on the basis of race,” id., at 916 (emphasis added). Caution is especially appropriate in this case, where the State has articulated a legitimate political explanation for its district-ing decision, and the voting population is one in which race and political affiliation are highly correlated. See Cromar-tie, supra, at 551-552 (noting that “[ejvidence that blacks constitute even a supermajority in one congressional district while amounting to less than a plurality in a neighboring district will not, by itself, suffice to prove that a jurisdiction was motivated by race in drawing its district lines when the evidence also shows a high correlation between race and party preference”).
We also are aware that we review the District Court’s findings only for “clear error.” In applying this standard, we, like any reviewing court, will not reverse a lower court’s finding of fact simply because we “would have decided the case differently.” Anderson v. Bessemer City, 470 U.S. 564, 578 (1985). Rather, a reviewing court must ask whether, “on the entire evidence,” it is “left with the definite and firm conviction that a mistake has been committed.” United States v. United States Gypsum Co., 838 U.S. 364, 395 (1948).
Where an intermediate court reviews, and affirms, a trial court’s factual findings, this Court will not “lightly overturn” the concurrent findings of the two lower courts. E. g., Neil [243] v. Biggers, 409 U.S. 188, 193, n. 3 (1972). But in this instance there is no intermediate court, and we are the only court of review. Moreover, the trial here at issue was not lengthy and the key evidence consisted primarily of documents and expert testimony. Credibility evaluations played a minor role. Accordingly, we find that an extensive review of the District Court’s findings, for clear error, is warranted. See Bose Corp. v. Consumers Union of United States, Inc., 466 U.S. 485, 500-501 (1984). That review leaves us “with the definite and firm conviction,” United States Gypsum Co., supra, at 395, that the District Court’s key findings are mistaken.
Ill
The critical District Court determination — the matter for which we remanded this litigation — consists of the finding that race rather than politics predominantly explains District 12’s 1997 boundaries. That determination rests upon three findings (the district’s shape, its splitting of towns and counties, and its high African-American voting population) that we previously found insufficient to support summary judgment. Cromartie, 526 U.S., at 547-549. Given the undisputed evidence that racial identification is highly correlated with political affiliation in North Carolina, these facts in and of themselves cannot, as a matter of law, support the District Court’s judgment. See Vera, 517 U.S., at 968 (O’Connor, J., principal opinion) (“If district lines merely correlate with race because they are drawn on the basis of political affiliation, which correlates with race, there is no racial classification to justify”). The District Court rested, however, upon five new subsidiary findings to conclude that District 12’s lines are the product of no “mer[e] correlation],” ibid., but are instead a result of the predominance of race in the legislature’s line-drawing process. See supra, at 240-241.
In considering each subsidiary finding, we have given weight to the fact that the District Court was familiar with [244] this litigation, heard the testimony of each witness, and considered all the evidence with care. Nonetheless, we cannot accept the District Court’s findings as adequate for reasons which we shall spell out in detail and which we can summarize as follows:
First, the primary evidence upon which the District Court relied for its “race, not politics,” conclusion is evidence of voting registration, not voting behavior; and that is precisely the kind of evidence that we said was inadequate the last time this case was before us. See infra, at 245-246. Second, the additional evidence to which appellees’ expert, Dr. Weber, pointed, and the statements made by Senator Cooper and Gerry Cohen, simply do not provide significant additional support for the District Court’s conclusion. See infra, at 246-250, 258-254. Third, the District Court, while not accepting the contrary conclusion of appellants’ expert, Dr. Peterson, did not (and as far as the record reveals, could not) reject much of the significant supporting factual information he provided. See infra, at 251-253. Fourth, in any event, appellees themselves have provided us with charts summarizing evidence of voting behavior and those charts tend to refute the court’s “race, not politics,” conclusion. See infra, at 254-257; Appendixes, infra.
A
The District Court primarily based its “race, not politics,” conclusion upon its finding that “the legislators excluded many heavily-Democratie precincts from District 12, even when those precincts immediately border the Twelfth and would have established a far more compact district.” 133 F. Supp. 2d, at 419; see also id., at 420 (“[M]ore heavily Democratic precincts ... were bypassed_in favor of precincts with a higher African-American population”). This finding, however — insofar as it differs from the remaining four— rests solely upon evidence that the legislature excluded heavily white precincts with high Democratic Party registra[245] tion, while including heavily African-American precincts with equivalent, or lower, Democratic Party registration. See id., at 413-414,415. Indeed, the District Court cites at length figures showing that the legislature included “several precincts with racial compositions of 40 to 100 percent African-American,” while excluding certain adjacent precincts “with less than 35 percent African-American population” but which contain between 54% and 76% registered Democrats. Id., at 414.
As we said before, the problem with this evidence is that it focuses upon party registration, not upon voting behavior. And we previously found the same evidence, compare ibid. (District Court’s opinion after trial) with App. to Juris. Statement 249a-250a (District Court’s summary judgment opinion), inadequate because registration figures do not accurately predict preference at the polls. See id., at 174a; see also Cromartie, supra, at 550-551 (describing Dr. Peterson’s analysis as “more thorough” because in North Carolina, “party registration and party preference do not always correspond”). In part this is because white voters registered as Democrats “cross-over” to vote for a Republican candidate more often than do African-Americans, who register and vote Democratic between 95% and 97% of the time. See Record, Deposition of Gerry Cohen 37-42 (discussing data); App. 304 (stating that white voters east about 60% to 70% of their votes for Republican candidates); id., at 139 (Dr. Weber’s testimony that 95% to 97% of African-Americans register and vote as Democrats); see also id., at 118 (testimony by Dr. Weber that registration data were the least reliable information upon which to predict voter behavior). A legislature trying to secure a safe Democratic seat is interested in Democratic voting behavior. Hence, a legislature may, by placing reliable Democratic precincts within a district without regard to race, end up with a district containing more heavily African-American precincts, but the reasons would be political rather than racial.
[246] Insofar as the District Court relied upon voting registration data, particularly data that were previously before us, it tells us nothing new; and the data do not help answer the question posed when we previously remanded this litigation. Cromartie, 526 U.S., at 551.
B
The District Court wrote that “[additionally, [plaintiffs’ expert, Dr. Weber, showed time and again how race trumped party affiliation in the construction of the 12th District and how political explanations utterly failed to explain the composition of the district.” 133 F. Supp. 2d, at 419. In support of this conclusion, the court relied upon six different citations to Dr. Weber’s trial testimony. We have examined each reference.
At the first cited pages of the trial transcript, Dr. Weber says that a reliably Democratic voting population of 60% is sufficient to create a safe Democratic seat. App. 91. Yet, he adds, the legislature created a more-than-60% reliable Democratic voting population in District 12. Hence (we read Dr. Weber to infer), the legislature likely was driven by race, not politics. Tr. 163; App. 314-315.
The record indicates, however, that, although Dr. Weber is right that District 12 is more than 60% reliably Democratic, it exceeds that figure by very little. Nor did Dr. Weber ask whether other districts, unchallenged by appellees, were significantly less "safe” than was District 12. Id., at 148. In fact, the figures the legislature used showed that District 12 would be 63% reliably Democratic. App. to Juris. Statement 80a (Democratic vote over three representative elections averaged 63%). By the same measures, at least two Republican districts (Districts 6 and 10) are 61% reliably Republican. Ibid. And, as Dr. Weber conceded, incumbents might have urged legislators (trying to maintain a [247] six/six Democrat/Republican delegation split) to make their seats, not 60% safe, but as safe as possible. App. 149. In a field such as voting behavior, where figures are inherently uncertain, Dr. Weber’s tiny calculated percentage differences are simply too small to carry significant evidentiaiy weight.
The District Court cited two parts of the transcript where Dr. Weber testified about a table he had prepared listing all precincts in the six counties, portions of which make up District 12. Tr. 204-205, 262. Dr. Weber said that District 12 contains between 39% and 56% of the precincts (depending on the county) that are more-than-40% reliably Democratic, but it contains almost every precinct with more-than-40% African-American voters. Id., at 204-205. Why, he essentially asks, if the legislature had had polities primarily in mind, would its effort to place reliably Democratic precincts within District 12 not have produced a greater racial mixture?
Dr. Weber’s own testimony provides an answer to this question. As Dr. Weber agreed, the precincts listed in the table were at least 40% reliably Democratic, but virtually all the African-American precincts included in District 12 were wore than 40% reliably Democratic. Moreover, none of the excluded white preeinets were as reliably Democratic as the African-American precincts that were included in the district. App. 140. Yet the legislature sought precincts that were reliably Democratic, not precincts that were 40% reliably Democratic, for obvious political reasons.
Neither does the table specify whether the excluded white-reliably-Democratie precincts were located near enough to District 12’s boundaries or each other for the legislature as a practical matter to have drawn District 12’s boundaries to have included them, without sacrificing other important political goals. The contrary is suggested by the [248] fact that Dr. Weber’s own proposed alternative plan, see id., at 106-107, would have pitted two incumbents against each other (Sue Myriek, a Republican from former District 9 and Mel Watt, a Democrat from former District 12). Dr. Weber testified that such a result — “a very competitive race with one of them losing their seat” — was desirable. Id., at 153. But the legislature, for political, not racial, reasons, believed the opposite. And it drew its plan to protect incumbents— a legitimate political goal recognized by the District Court. 133 F. Supp. 2d, at 412-413.
For these reasons, Dr. Weber’s table offers little insight into the legislature’s true motive.
The next part of the transcript the District Court cited contains Dr. Weber’s testimony about a Mecklenburg County precinct (precinct 77) which the legislature split between Districts 9 and 12. Tr. 221. Dr. Weber apparently thought that the legislature did not have to split this precinct, placing the more heavily African-American segment within District 12 — unless, of course, its motive was racial rather than political. But Dr. Weber simultaneously conceded that he had not considered whether District 9’s incumbent Republican would have wanted the whole of precinct 77 left in her own district where it would have burdened her with a significant additional number of reliably Democratic voters. App. 156-157. Nor had Dr. Weber “test[ed]” his conclusion that this split helped to show a racial (rather than political) motive, say, by adjusting other boundary lines and determining the political, or other nonraeial, consequences of such adjustments. Id., at 132.
The maps in evidence indicate that to have placed all of precinct 77 within District 12 would have created a District 12 peninsula that invaded District 9, neatly dividing that latter district in two, see id., at 496 — a conclusive nonracial reason for the legislature’s decision not to do so.
[249]*2494
The District Court cited Dr. Weber’s conclusion that “race is the predominant factor.” Tr. 251. But this statement of the conclusion is no stronger than the evidence that underlies it.
The District Court’s final citation is to Dr. Weber’s assertion that there are other ways in which the legislature could have created a safely Democratic district without placing so many primarily African-American districts within District 12. Id., at 288. And we recognize that some such other ways may exist. But, unless the evidence also shows that these hypothetical alternative districts would have better satisfied the legislature’s other nonraeial political goals as well as traditional nonraeial districting principles, this fact alone cannot show an improper legislative motive. After all, the Constitution does not place an affirmative obligation upon the legislature to avoid creating districts that turn out to be heavily, even majority, minority. It simply imposes an obligation not to create such districts for predominantly racial, as opposed to political or traditional, districting motivations. And Dr. Weber’s testimony does not, at the pages cited, provide evidence of a politically, practical alternative plan that the legislature failed to adopt predominantly for racial reasons.
In addition, we have read the whole of Dr. Weber’s testimony, including portions not cited by the District Court. Some of those portions further undercut Dr. Weber’s conclusions. Dr. Weber said, for example, that he had developed those conclusions while under the erroneous impression that the legislature’s computer-based districting program provided information about racial, but not political, balance. App. 187-138; see also id., at 302 (reflecting Dr. Weber’s erroneous impression in the declaration he submitted to [250] the District Court). He also said he was not aware of “anything about political dynamics going on in the [legislature involving” District 12, id., at 135, sometimes expressing disdain for a process that we have cautioned courts to respect, id., at 150-151; Miller, 515 U.S., at 915-916.
Other portions support Dr. Weber’s conclusions. Dr. Weber testified, for example, about a different alternative plan that, in his view, would have provided both greater racial balance and political security, namely, a plan that the legislature did enact in 1998, and which has been in effect during the time the courts have been reviewing the constitutionality of the 1997 plan, App. 156-157. The existence of this alternative plan, however, cannot help appellees significantly. Although it created a somewhat more compact district, it still divides many communities along racial lines, while providing fewer reliably Democratic District 12 voters and transferring a group of highly Democratic precincts into two safely Republican districts, namely, the 5th and 6th Districts, which political result the 1997 plan sought to avoid. See Tr. 352, 355. Furthermore, the 1997 plan before this Court, unlike the 1998 plan, joined three major cities in a manner legislators regarded as reflecting “a real commonality of urban interests, with inner city schools, urban health care . . . problems, public housing problems.” App. 430 (statement of Sen. Winner); see also id., at 421 (statement of Sen. Martin). Consequently, we cannot tell whether the existence of the 1998 plan shows that the 1997 plan was drawn with racial considerations predominant. And, in any event, the District Court did not rely upon the existence of the 1998 plan to support its ultimate conclusion. See Kelley v. Everglades Drainage Dist, 319 U.S. 415, 420-422 (1943) (per curiam).
We do not see how Dr. Weber’s testimony, taken as a whole, could have provided more than minimal support for the District Court’s conclusion that race predominantly underlay the legislature’s districting decision.
[251] C
The District Court found that the testimony of the State’s primary expert, Dr. Peterson, was “'unreliable’ and not relevant.” 188 F. Supp. 2d, at 420 (quoting Dr. Weber and citing Tr. 222-224, 232). Dr. Peterson’s testimony was designed to show that African-American Democratic voters were more reliably Democratic and that District 12’s boundaries were drawn to include reliable Democrats. Specifically, Dr. Peterson compared precincts immediately within District 12 and those immediately without to determine whether the boundaries of the district corresponded better with race than with politics. The principle underlying Dr. Peterson’s analysis is that if the district were drawn with race predominantly in mind, one would expect the boundaries of the district to correlate with race more than with politics.
The pages cited in support of the District Court’s rejection of Dr. Peterson’s conclusions contain testimony by Dr. Weber, who says that Dr. Peterson’s analysis is unreliable because (1) it “ignor[es] the core” of the district, id., at 223, and (2) it fails to take account of the fact that different precincts have different populations, id., at 223-224. The first matter — ignoring the “core” — apparently reflects Dr. Weber’s view that in context the fact that District 12’s heart or “core” is heavily African-American by itself shows that the legislature’s motive was predominantly racial, not political. The District Court did not argue that the racial makeup of a district’s “core” is critical. Nor do we see why “core” makeup alone could help the court discern the relevant legislative motive. Nothing here suggests that only “core” makeup could answer the “political/racM” question that this Court previously found critical. Cromartie, 526 U. S., at 551-552.
The second matter — that Dr. Peterson’s boundary segment analysis did not account for differences in population between precincts — relates to one aspect of Dr. Peterson’s [252] testimony., Appellants presented Dr. Peterson’s testimony and data in support of four propositions: first, that registration figures do not accurately reflect actual voting behavior, see App. to Juris. Statement 178a-174a; second, that African-Americans are more reliable Democrats than whites, see id., at 159a-160a; third, that political affiliation explains splitting cities and counties as well as does race, see id., at 189a, 191a-192a, 182a-185a; and fourth, that differences in the racial and political makeup of the precincts just inside and outside the boundaries of District 12 show that polities is as good an explanation as is race for the district’s boundaries, see id., at 161a~167a; 181a-182a. The District Court’s criticism of Dr. Peterson’s testimony at most affects the reliability of the fourth element of Dr. Peterson’s testimony, his special boundary segment analysis. The District Court’s criticism of Dr. Peterson’s boundary segment analysis does not undermine the data related to the split communities. The criticism does not undercut Dr. Peterson’s presentation of statistical evidence showing that registration was a poor indicator of party preference and that African-Americans are much more reliably Democratic voters, nor have we found in the record any significant evidence refuting that data.
At the same time, appellees themselves have used the information available in the record to create maps comparing the district’s boundaries with Demoeratie/Republican voting behavior. See Appendixes A, B, and C, infra. Because no one challenges the accuracy of these maps, we assume that they are reliable; and we can assume that Dr. Peterson’s testimony is reliable insofar as it confirms what the maps themselves contain and appellees themselves concede. Those maps, with certain exceptions discussed below, see infra, at 254-257, further indicate that the legislature drew boundaries that, in general, placed more-reliably Democratic voters inside the district, while placing less-reliably Democratic voters outside the district. And that fact, in turn, [253] supports the State’s answers to the questions we previously found critical.
D
The District Court also relied on two pieces of “direct” evidence of discriminatory intent.
The court found that a legislative redistricting leader, Senator Roy Cooper, when testifying before a legislative committee in 1997, had said that the 1997 plan satisfies a “need for ‘racial and partisan’ balance.” 133 F. Supp. 2d, at 419. The court concluded that the words “racial balance” referred to a 10-to-2 Caucasian/African-American balance in the State’s 12-member congressional delegation. Ibid. Hence, Senator Cooper had admitted that the legislature had drawn the plan with race in mind.
Senator Cooper’s full statement reads as follows:
“Those of you who dealt with Redistricting before realize that you cannot solve each problem that you encounter and everyone can find a problem with this Plan. However, I think that overall it provides for a fair, geographic, racial and partisan balance throughout the State of North Carolina. I think in order to come to an agreement all sides had to give a little bit, but I think we’ve reached an agreement that we can live with.” App. 460.
We agree that one can read the statement about “racial... balance” as the District Court read it — to refer to the current congressional delegation’s racial balance. But even as so read, the phrase shows that the legislature considered race, along with other partisan and geographic considerations; and as so read it says little or nothing about whether race played a predominant role comparatively speaking. See Vera, 517 U.S., at 958 (O’Connor, J., principal opinion) (“Strict scrutiny does not apply merely because redistricting [254] is performed with consciousness of race”); see also Miller, 515 U.S., at 916 (legislatures “will . . . almost always be aware of racial demographics”); Shaw I, 509 U. S., at 646 (same).
The second piece of “direct” evidence relied upon by the District Court is a February 10, 1997, e-mail sent from Gerry Cohen, a legislative staff member responsible for drafting districting plans, to Senator Cooper and Senator Leslie Winner. Cohen wrote: “I have moved Greensboro Black community into the 12th, and now need to take [about] 60,000 out of the 12th. I await your direction on this.” App. 369.
The reference to race — i. e., “Black community” — is obvious. But the e-mail does not discuss the point of the reference. It does not discuss why Greensboro’s African-American voters were placed in the 12th District; it does not discuss the political consequences of failing to do so; it is addressed only to two members of the legislature; and it suggests that the legislature paid less attention to raee in respect to the 12th District than in respect to the 1st District, where the e-mail provides a far more extensive, detailed discussion of racial percentages. It is less persuasive than the kinds of direct evidence we have found significant in other redistricting cases. See Vera, supra, at 959 (O’Con-nor, J., principal opinion) (State conceded that one of its goals was to create a majority-minority district); Miller, supra, at 907 (State set out to create majority-minority district); Shaw II, 517 U. S., at 906 (recounting testimony by Cohen that creating a majority-minority district was the “principal reason” for the 1992 version of District 12). Nonetheless, the e-mail offers some support for the District Court’s conclusion.
E
As we have said, we assume that the maps appended to appellees’ brief reflect the record insofar as that record [255] describes the relation between District 12’s boundaries and reliably Democratic voting behavior. Consequently we shall consider appellees’ related claims, made on appeal, that the maps provide significant support for the District Court, in that they show how the legislature might have “swapped” several more heavily African-American District 12 precincts for other less heavily African-American adjacent precincts— without harming its basic “safely Democratic” political objective. Cf. supra, at 246-247.
First, appellees suggest, without identifying any specific swap, that the legislature could have brought within District 12 several reliably Democratic, primarily white, precincts in Forsyth County. See Brief for Appellees 30. None of these precincts, however, is more reliably Democratic than the precincts immediately adjacent and within District 12. See Appendix A, infra (showing Democratic strength reflected by Republican victories in each precinct); App. 484 (showing Democratic strength reflected by Democratic registration). One of them, the Brown/Douglas Recreation Precinct, is heavily African-American. See ibid. And the remainder form a buffer between the home precinct of Fifth District Representative Richard Burr and the District 12 border, such that their removal from District 5 would deprive Representative Burr of a large portion of his own hometown, making him more vulnerable to a challenge from elsewhere within his district. App. to Juris. Statement 209a; App. 623. Consequently the Forsyth County precincts do not significantly help appellees’ “race, not politics,” thesis.
Second, appellees say that the legislature might have swapped two District 12 Davidson County precincts (Thom-asville 1 and Lexington 3) for a District 6 Guilford County preeinct (Greensboro 17). See Brief for Appellees 30, n. 25. Whatever the virtues of such a swap, however, it would have diminished the size of District 12, geographically producing an unusually narrow isthmus linking District 12’s north with its south and demographieally producing the State’s smallest [256] district, deviating by about 1,300 below the legislatively endorsed ideal mean of 552,386 population. Traditional dis-tricting considerations consequently militated against any such swap. See Record, Deposition of Linwood Lee Jones 122 (stating that legislature’s goal was to keep deviations from ideal population to less than 1,000); App. 199 (testimony of Sen. Cooper to same effect).
Third, appellees suggest that, in Mecklenburg County, two District 12 precincts (Charlotte 81 and LCI-South) be swapped with two District 9 precincts (Charlotte 10 and 21). See Brief for Appellees 30, n. 25. This suggestion is difficult to evaluate, as the parties provide no map that specifically identifies each precinct in Mecklenburg County by name. Nonetheless, from what we can tell, such a swap would make the district marginally more white (decreasing the African-American population by about 300 persons) while making the shape more questionable, leaving the precinct immediately to the south of Charlotte 81 jutting out into District 9. We are not convinced that this proposal materially advances appellees’ claim.
Fourth, appellees argue that the legislature could have swapped two reliably Democratic Greensboro precincts outside District 12 (11 and 14) for four reliably Republican High Point precincts (1, 13,15, and 19) placed within District 12. See ibid. The swap would not have improved racial balance significantly, however, for each of the six precincts have an African-American population of less than 35%. Additionally, it too would have altered the shape of District 12 for the worse. See Appendix D, infra; see also App. 622 (testimony of Gerry Cohen). And, in any event, the decision to exclude the two Greensboro precincts seems to reflect the legislature’s decision to draw boundaries that follow main thoroughfares in Guilford County. App. to Juris. Statement 205a; App. 575.
Even if our judgments in respect to a few of these precincts are wrong, a showing that the legislature might [257] have “swapped” a handful of precincts out of a total of 154 precincts, involving a population of a few hundred out'of a total population of about half a million, cannot significantly strengthen appellees’ case.
IV
We concede the record contains a modicum of evidence offering support for the District Court’s conclusion. That evidence includes the Cohen e-mail, Senator Cooper’s reference to “racial balance,” and to a minor degree, some aspects of Dr. Weber’s testimony. The evidence taken together, however, does not show that racial considerations predominated in the drawing of District 12’s boundaries. That is because race in this case correlates closely with political behavior. The basic question is whether the legislature drew District 12’s boundaries because of race rather than because of political behavior (coupled with traditional, nonracial districting considerations). It is not, as the dissent contends, see post, at 266 (opinion of Thomas, J.), whether a legislature may defend its districting decisions based on a “stereotype” about African-American voting behavior. And given the fact that the party attacking the legislature’s decision bears the burden of proving that racial considerations are “dominant and controlling,” Miller, 616 U. S., at 913, given the “demanding” nature of that burden of proof, id., at 929 (O’Connor, J., concurring), and given the sensitivity, the “extraordinary caution,” that district courts must show to avoid treading upon legislative prerogatives, id., at 916 (majority opinion), the attacking party has not successfully shown that race, rather than politics, predominantly accounts for the result. The record leaves us with the “definite and firm conviction,” United States Gypsum Co., 333 U.S., at 395, that the District Court erred in finding to the contrary. And we do not believe that providing appellees a further opportunity to make their “precinct swapping” arguments in the District Court could change this result.
[258] We can put the matter more generally as follows: In a ease such as this one where majority-minority districts (or the approximate equivalent) are at issue and where racial identification correlates highly with political affiliation, the party attacking the legislatively drawn boundaries must show at the least that the legislature could have achieved its legitimate political objectives in alternative ways that are comparably consistent with traditional districting principles. That party must also show that those districting alternatives would have brought about significantly greater racial balance. Appellees failed to make any such showing here. We conclude that the District Court’s contrary findings are clearly erroneous. Because of this disposition, we need not address appellants’ alternative grounds for reversal.
The judgment of the District Court is
Reversed.
[Appendixes containing maps from appellees’ and appellants’ briefs follow this page.]
[0] [[Image here]]
[[Image here]]
[[Image here]]
[[Image here]]