dissenting:
Arizona has criminalized one of the most popular and effective methods by which minority voters cast their ballots. Because this law violates the Constitution and the Voting Rights Act, I must respectfully dissent.
I
Like most states, Arizona allows voters to cast a ballot on election day at a polling place, or to cast an early absentee vote, either in person or by mail. A.R.S. § 16-541. Early voting has become increasingly popular in Arizona, as evidenced by the fact that 81% of ballots cast in the last Presidential election were cast by early voting, a 12% increase from the 2012 election. An important reason for the increase in early voting is that Arizona has substantially reduced the number of polling places, resulting in extraordinarily long lines, with voters waiting many hours to cast their ballots. In one urban area, there is one voting center for nearly 70,000 registered voters. In some precincts in Maricopa County, voters waited for four hours to cast their ballots in the Presidential Preference Primary election earlier this year. In other precincts, the wait was up to six hours. Compounding the problem is that, in Maricopa County in particular, polling places change with each election, and the County is using a different polling place system for the general election than it did in the Presidential Preference election earlier this year.
As the use of early voting has skyrocketed in Arizona, voters have increasingly used friends, organizations, .political parties, and campaign workers to transmit their ballots. Some efforts are typical of “get-out-the-vote” campaigns by partisan groups; others are targeted to provide a service to those who cannot get to the polls. Because of geographic and other impediments to voting, voting by ballot collection has become a critical means for minority voters to cast their ballots. A substantial number of rural minority voters live in areas without easy access to mail service. In urban areas, many minority voters are socioeconomically disadvantaged, meaning that they may lack reliable mail service and have to rely on public transportation to get to polling places.
Nonetheless, Arizona enacted the law at issue, House Bill 2023, codified at A.R.S. § 16-1005 (H)-(I), which imposes felony criminal sanctions for non-household members or caregivers who collect early ballots from others. Plaintiffs filed this lawsuit challenging the law under the Voting Rights Act of 1965 and the First and Four*1085teenth Amendments to the United States Constitution. The district court denied the plaintiffs’ motion for a preliminary injunction, and this interlocutory appeal followed.
We review the denial of a preliminary injunction for abuse of discretion. All. for the Wild Rockies v. Cottrell, 632 F.3d 1127, 1131 (9th Cir. 2011). A district court abuses its discretion if its analysis is premised on an inaccurate view of the law. Pom Wonderful LLC v. Hubbard, 775 F.3d 1118, 1123 (9th Cir. 2014). In such instances, we review de novo the legal premises underlying the preliminary injunction. Id.1
II
The district court erred in its analysis of the plaintiffs’ Fourteenth Amendment claims. First, it erroneously employed a rational basis review standard, when the appropriate standard was a “balancing and means-end fit analysis.” Pub. Integrity All. v. City of Tucson, 836 F.3d 1019, 1025 (9th Cir. 2016) (en banc). As Public Integrity Alliance recognized, the Supreme Court established the, appropriate standard of review for laws regulating the right to vote in Burdick v. Takushi, 504 U.S. 428, 112 S.Ct. 2059, 119 L.Ed.2d 245 (1992). As we explained in Public Integrity Alliance:
Under Burdick’s balancing, and means-end fit framework, strict scrutiny is appropriate when First or Fourteenth Amendment rights “are subjected to ‘severe’ restrictions.” Id. (quoting Norman v. Reed, 502 U.S. 279, 289, 112 S.Ct. 698, 116 L.Ed.2d 711 (1992)). “But when a state election law provision imposes only ‘reasonable, nondiscriminatory restrictions’ upon the First and Fourteenth Amendment rights of voters, ‘the State’s important regulatory interests are generally sufficient to justify’ the restrictions.” Id. (quoting Anderson, 460 U.S. at 788, 103 S.Ct. 1564).
Pub. Integrity All., 836 F.3d at 1023.
However, rather than reviewing H.B, 2023 under a balancing and means-end fit analysis, the district court conducted a rational basis review, committing legal error.2
*1086The second, and more important legal error, was that the district court misapplied the analysis required by Burdick and Anderson v. Celebrezze, 460 U.S. 780, 789, 103 S.Ct. 1564, 75 L.Ed.2d 547 (1983). Under Andersón-Burdick, the court must weigh the nature and magnitude of the burden imposed by the law against the state’s interest and justification for it. Nader v. Brewer, 531 F.3d 1028, 1034 (9th Cir. 2008).
The burden of the law on Arizona minority voters is substantial and occurs in both urban and rural areas of the state. The uncontradicted evidence presented to the district court showed that a substantial number of minority voters used ballot collection as their means of voting. As Mari-copa Board of Supervisors Steve Gallardo testified: “ballot collectors are used in large part by Latino and Native American groups and [ballot collecting] has come to be critical in enabling voters in those communities to exercise their fundamental right to vote.”
The record demonstrated that, in many rural areas with a high proportion of minority voters, home mail delivery was not available, and it was extremely difficult to travel to a post office. No one contested the fact that the rural communities of Somerton and San Luis, which are comprised of 95.9% and 98.7% Hispanic voters, respectively, were without home mail delivery and reliable transportation. As the representative for that district testified, “[bjecause many of these voters are elderly and have mobility challenges, it is a common practice in this area to have one neighbor pick up and drop off mail for others on their street as a neighborly service.” The representative noted that there is only one post office, which is- located across a highway crowded with cars waiting to cross the border, and is virtually inaccessible by foot.
Another example of the impact of the law on minority voters is the Tohono O’od-ham Indian Nation. The Tohono O’odham reservation constitutes over 2.8 million acres in the Sonoran desert. It is an area larger than Rhode Island and Delaware, and approximates the size of Connecticut. It has about 14,000 registered voters. It does not have home mail delivery. It has one post office, which is over 40 miles away from many residents. The evidence in this case shows that restrictions on ballot collection affect the Tohono O’odham tribe significantly. No one contested the fact that the members of the Tohono O’od-ham Indian Nation have limited access to a postal service and no home mail delivery.
Similarly, no one disputed that members of the Cocopah Indian Tribe do not have home mail delivery or easy access to a post office. The Cocopah Reservation is located along the lower Colorado River, south of Yuma, Arizona. The Cocopah Reservation comprises approximately 6,500 acres, with approximately 1,000 tribal members who live and work on or near the Reservation.
As to urban areas, record evidence demonstrated that the burden of the law affected minority voters the most because of socioeconomic factors. Minority voters in urban areas were more likely to be economically disadvantaged. The record showed that many minority urban voters lived in places with insecure mail delivery; that many minority urban voters were de*1087pendent upon public transportation, which made election day in-person voting difficult; that many minority voters worked several jobs, making it difficult to take time off work to vote in person; and that many infirm minority voters did not have access to caregivers or family who could transmit ballots.
Martin Quezada, State Senator for Arizona’s Twenty-Ninth Senate District testified that:
I represent approximately 213,000 constituents, nearly 80% of which are ethnic minorities. In particular, Hispanic citizens comprise 67% of the population of my district, the highest percentage of any district in the state of Arizona. My district is a working-class community, and many of my constituents depend on public transportation. Many of my constituents were severely burdened by the long lines and lack of polling locations in the 2016 presidential preference election. My entire district only had one vote center, in Maryvale, to service the nearly 70,000 registered voters.
The President of a nonprofit organization comprised of Latino citizens and community leaders testified that many minorities required assistance in making sure that they were following the proper voting procedure, and in low income areas they were concerned about the security of their mailboxes.
Further complicating voting in Arizona’s urban areas is that there are not only few places to vote, but that the polling locations change frequently. Indeed, because the City of Phoenix elections are run independently by the City, a voter might have to go to two different polling places to cast ballots on election day. According -to the Executive Director of a nonprofit organization working primarily in low-income African-American and Latino, neighborhoods, this confusion significantly burdened those communities because many minorities had difficulty navigating the voting process, especially those Spanish-speaking voters who were not also fluent in English. The record also showed that election administrators were prone to make errors with Spanish-language materials. Those voters encounter significant hurdles at polling places. Thus, the opportunity for early voting is especially important for those citizens. ■
The district court and the State dismiss the burdens imposed on minority voters seéking to vote early as attacks on a process that provides only ■ a “more' convenient” means of voting." However, when 80% of the electorate uses early absentee voting as the method by which they cast their ballots, the method has transcended convenience and has become instead a practical necessity. Thus, when severe burdens are placed on this form of voting, it has a significant impact on elections and the right to vote.
Against this burden, the state’s justification for the law was weak; The state identified" its interest as preventing voter fraud. However, the sponsors of the legislation could not identify a single example of voter fraud caused by ballot collection. Not one. Nor is there a single example in the record of this case. The primary proponent of the legislation admitted there were no examples of such fraud, but that the legislation was ba,sed on the speculative theory that fraud could occur. A study by the Arizona Republic found that, out of millions of ballots cast from 2005 to 2013, there were only 34 cases of fraud prosecution. All involved voting by felons or non-citizens. None involved any allegation of fraud in "ballot collection. And none of'the cases resulted in a conviction. A study by the National Republican Lawyers Association,- which was dedicated to finding voter fraud "and investigated evidence of potential fraud between 2000 and 2011, uncovered no example of fraud resulting from *1088the collection and delivery of early ballots in Arizona. A follow-up analysis through May of 2015 failed to uncover any examples of ballot collection fraud. The plaintiffs produced numerous affidavits that attested that no one associated with ballot collection had ever witnessed any voter fraud. Further, the record indicated that there are effective processes in place to handle any ballot that exhibits any signs that tampering has occurred. The Director of Elections for Maricopa County, the most populated county in Arizona, with a population of four times the next most populated county, testified at the legislative hearings that the County was well equipped to deal with voter fraud. Under state election procedure, voters can check the status of their ballot at any time. In short, the specter of voter fraud by ballot collection is much like the vaunted opening of A1 Capone’s vault: there is simply nothing there.
Thus, when one balances the serious burdens placed on minorities by the law against the extremely weak justification offered by the state, one can only conclude under the Anderson-Burdick analysis that the plaintiffs have established a likelihood of success on the merits of their Fourteenth Amendment claim.3 Based on the mostly uncontroverted record, the district court erred in misapplying Anderson-Burdick.4
III
The district court also erred in denying the motion for a preliminary injunction based on the Voting Rights Act claims. The Voting Rights Act of 1965 “was designed by Congress to banish the blight of racial discrimination in voting, which has infected the electoral process in parts of our country for nearly a century.” State of S.C. v. Katzenbach, 383 U.S. 301, 308, 86 S.Ct. 803, 15 L.Ed.2d 769 (1966) abrogated by Shelby Cty., Ala. v. Holder, — U.S. -, 133 S.Ct. 2612, 186 L.Ed.2d 651 (2013). The Act “implemented Congress’ firm intention to rid the country of racial discrimination in voting. It provided stringent new remedies against those practices which have most frequently denied citizens the right to vote on the basis of their race.” Allen v. State Bd. of Elections, 393 *1089U.S. 544, 548, 89 S.Ct. 817, 22 L.Ed.2d 1 (1969).
The central purpose of the Act was “[t]o enforce the fifteenth amendment to the Constitution of the United States.” Chisom v. Roemer, 501 U.S. 380, 383, 111 S.Ct. 2354, 115 L.Ed.2d 348 (1991) (quoting Pub.L. 89-110, 79 Stat. 437, 42 U.S.C. § 1973 et seq.). The Fifteenth Amendment provides that “[t]he right of citizens of the United States to vote shall not be denied or abridged by the United States or by any State on account of race, color, or previous condition of servitude.” U.S. Const, amend. XV, § 1.
At issue in this case is § 2 of the Act, which is “a restatement of the Fifteenth Amendment.” Roemer, 501 U.S. at 392, 111 S.Ct. 2354. Section 2 provides, without limitation, that any voting qualification that denies citizens the right to vote in a discriminatory manner violates the Voting Rights Act. 42 U.S.C. § 1973; see also Allen, 393 U.S. at 566-67, 89 S.Ct. 817 (noting that Congress intentionally chose the expansive language “voting qualifications or prerequisite to voting, or standard, practice, or procedure” for § 2 so as to be “all-inclusive of any kind of practice” that might be used by states to deny citizens the right to vote (infernal quotation marks omitted)). As amended in 1982, § 2 makes “clear that certain practices and procedures that result in the denial or abridgment of the right to vote are forbidden even though the absence of proof , of discriminatory intent protects them from constitutional challenge.” Roemer, 501 U.S. at 383-84, 111 S.Ct. 2354.
To succeed on a § 2 claim, a plaintiff must show (1) that “the challenged standard, practice, or procedure must impose a discriminatory burden on members of a protected class, meaning that members of the protected class have less opportunity than other members of the electorate to participate in the political process and to elect representatives of their choice” and (2) “that burden must in part be caused by or linked to social and historical conditions that have or currently produce discrimination against members of the protected class.” League of Women Voters of N.C. v. North Carolina, 769 F.3d 224, 240 (4th Cir. 2014) (internal quotations omitted); see also Veasey v. Abbott, 830 F.3d 216, 244 (5th Cir. 2016).
The district court made a number of legal errors in its analysis of the § 2 claims, warranting reversal.
A
The district court erred in holding, as a matter of law, that § 2 requires proof of the disparate impact of a law by “quantitative or statistical evidence comparing the proportion of minority versus white voters who rely on others to collect their early ballots.” As the State concedes, there is no case law supporting this requirement; the district court relied only on cases it thought “strongly suggested” it.
Although quantitative or statistical measures of comparing minority and white voting patterns certainly may provide important analytic evidence, the district court erred in concluding that they were the exclusive means of proof. Indeed, the district court’s conclusion is belied by the words of the Voting Rights Act itself, which provides that a violation of § 2 is “based oyi the totality of the circumstances.” 52 U.S.C. § 10301(b) (emphasis added). The statute requires evidence that members of the affected minority class “have less opportunity than other members of the electorate to participate in the political process and to elect representatives of their choice.” Id. (emphasis added). The statutory criterion is incompatible with the district court’s restriction of proof to quantitative denial of actual minority voting compared with white voting. The *1090relevant question is whether the challenged practice, viewed in the totality of the circumstances, places a disproportionate burden on the opportunities of minorities to vote. Veasy, 830 F.3d at 244-45; League of Women Voters, 769 F.3d at 240. Even when analyzing the second part of the § 2 test, which does require causality, statistical analyses are not the exclusive method of showing a violation.5 Veasy, 830 F.3d at 244. Indeed, the Supreme Court has eschewed that approach in favor of consideration of various factors. Gingles, 478 U.S. at 44-45, 106 S.Ct. 2752. Rather than narrowly interpreting the Voting Rights Act, the Supreme Court has emphasized its “broad remedial purpose of rid[ding] the country of racial discrimination in'voting” and has explained that it provided “the broadest possible scope in combating racial discrimination.” Roemer, 501 U.S. at 403, 111 S.Ct. 2354. The district court’s mechanical formulation is also at odds with the “totality of the circumstances” approach we underscored in Gonzalez v. Arizona, 677 F.3d 383, 406 (9th Cir. 2012). The district court’s restriction constitutes legal error.
• Even if we leave aside the irreconcilable conflict between the district court’s proposed rule and the requirements of the governing statute, the district’s approach is still fatally flawed;
First, quantitative measurement of the effect of a rule on the voting behavior of different demographic populations must necessarily occur after the election. One cannot statistically test the real world effect of a rule in the abstract; it can only be measured by actual voting data. In other words, imposition of the district court’s proposed rule would mean that there could never be a successful pre-election challenge of the burdens placed on minority voting opportunity because no data will have been generated or collected. The analysis could only occur after the harm had been inflicted. That result cannot be squared with the broad remedial purposes of the Voting Rights Act. The Fifth Circuit, in rejecting an approach similar to the district court’s, acknowledged this problem, observing that requiring such proof would “presentí ] problems for pre-election challenges ... when no such data is yet available.” Veasey, 830 F.3d at 260.
Second, the relevant data is not available in Arizona. The State concedes that it does not collect the necessary data, and asserts that it should not bear that burden in the absence of a law that requires it to do so. The State suggests that plaintiffs should use data from those organizations who collect ballots. Of course, that action would now be a felony. But leaving that aside, there would be no practical way for the plaintiffs to collect comparative data by that method because it is highly unlikely they could force competing organizational groups to collect and supply the data. And such a method would not likely yield true comparative results. At best, it would show *1091that white voters and minority voters both have used ballot collection as a means for casting their ballots. No one disputes that, nor does anyone seriously dispute the fact that minority citizens are especially dependent on ballot collection has a means of voting. Further, even if past data were available, it still would not,, answer the district court’s. query because the data gathered would be pre-rule, and therefore not relevant as a means of assessing the rule’s impact.
Third, the district court acknowledged the difficulty of obtaining the data because “election and other public records often do not include racial or ethnic data,” and noted that “[tjhere is no getting around this problem.” Nonetheless, the court held that the statute still required a threshold statistical showing, even though collecting such evidence was likely impossible. That was not the intent of the Voting Rights Act, and it is just such a circumstance that requires assessment of the “totality of the circumstances.”
Fourth, in its examination of the plaintiffs’ evidence, the district court erred in its comparative analysis. It faulted the plaintiffs for not showing comparative data from other rural white-centric areas. But that is not the examination required by the Voting Rights Act. Section 2 examines whether “members of the protected class have less opportunity than other members of the electorate to participate in the political process and to elect representatives of their choice.” Veasey, 830 F.3d at 305; League of Women Voters of N.C., 769 F.3d at 240 (emphasis added). It does not test opportunity against “other members of the electorate” who are “similarly situated.” Thus, contrary to the district court’s analysis, the comparison is not with similarly situated white groups, but rather with the voting population as a whole. If the district court’s assumption were correct, then literacy and poll tax statutes would be constitutional because they placed the burdens on illiterate and poor whites and blacks equally. Instead, the Voting Rights Act focuses on the burdens disproportionately place on minorities in comparison with the general voting population. Native American- voters living on reservations have different burdens as to transportation and mail access than urban white voters. A state may not evade the requirements of § 2 by arguing that it equally applies to a subset of white voters constituting a minuscule percentage of the white vote, when the overall effect is the suppression minority voting.
And even if we were to take the district court’s analysis at face value, it fails in consideration of the evidence in this case. The district court’s conclusion is at odds with the evidence showing the law disproportionately burdens minorities. I have previously described the situation faced by the Tohono O’odham Nation, situated on 2.8 million acres, with limited access to a post office and no home mail delivery. Everyone concedes that there is no white population analogue. There are no white reservations in Arizona. There is no comparably sized rural area that encompasses a white-majority population. The record evidence was plain and uncon-troverted: H.B. 2023 places a disproportionate burden on the voting opportunities of members of the Tohono O’odham tribe in comparison with the population of white voters.
The evidence provided by the plaintiffs showed a similar pattern in urban areas. Minority voters encountered significant burdens in exercising their right to vote. The reduced number of polling places meant that voters had to wait hours in line to cast ballots. Low income voters had difficulty getting to the polls because of their dependence on public transportation. Voters who were not.fluent in English had difficulty determining where to vote. Sta*1092tistical evidence is not needed to see that without ballot collecting, these voters will have less opportunity than other members of the electorate to participate in the political process.
In sum, the district court committed legal error by requiring the plaintiffs to show proof of the disparate impact of the law by “quantitative or statistical evidence comparing the proportion of minority versus white ■ voters who rely on others to collect their early ballots.”- That formulation is at odds with the governing statute, which requires- analysis by “totality of the circumstances” of whether members of the affected minority class “have less opportunity than other members of the electorate to participate in the political process and to elect representatives of their choice.” 52 U.S.C. § 10301(b).
B
The district court also erred as a matter of law in its assessment of the plaintiffs’ burden of proof. “[T]he burden of proof at the preliminary injunction phase tracks the burden of proof at trial .... ” Thalheimer v. City of San Diego, 645 F.3d 1109, 1116 (9th Cir. 2011). In a voting rights case, the plaintiff bears the burden of proof at trial and must' show a violation by a preponderance of the evidence. Bartlett v. Strickland, 556 U.S. 1, 19-20, 129 S.Ct. 1231, 173 L.Ed.2d 173 (2009). Thus, the parties seeking a preliminary injunction in this case must show they are likely to prevail on the merits; if the plaintiffs satisfy that burden, then the opposing parties bear the burden of rejoinder. Thalheimer, 645 F.3d at 1116.
Here, the district court rejected plaintiffs’ tendered evidence because it was not “compelling.” At the preliminary injunction stage, the plaintiff is not required to present “compelling” evidence, but only to establish a likelihood of success by a preponderance of the evidence. The district court also rejected the tendered evidence as “anecdotal,” but the Supreme Court has considered and credited just such evidence. At the preliminary injunction stage, plaintiffs were obligated to show a likelihood of success in showing that “members of the protected class have less opportunity than other members of the electorate to participate in the political process and to elect representatives of their choice.”
Much of the evidence tendered by the plaintiffs as to this burden was not controverted. As I have noted, no one contested the fact that the rural communities of Somerton and San Luis, which are comprised of 95.9% and 98.7% Hispanic voters, respectively, were without home mail delivery and reliable transportation. No one contested the fact that the members of the Tohono O’odham Indian Nation do not have home mail delivery. No one disputed that members of the Cocopah Indian Nation do not have home mail delivery. The plaintiffs submitted voluminous affidavits showing the burden that the restriction on ballot collection would impose on minorities. The State did not contest the affidavits, but simply dismissed the evidence as “anecdotal.” Thus, much of the evidence tendered by the plaintiffs as to the disproportionate burden on minority voters was either completely undisputed or uncontested.
However, once the plaintiffs had established the burden on minority voters, the district court did not place the burden of rejoinder on the State. Rather, it categorically rejected evidence based on personal knowledge as “anecdotal,” and held that the plaintiffs were required to show that rural white voters were not similarly burdened. In other words, once the plaintiffs had established the burden on minority voters, the district court imposed a higher standard of proof, rather than shifting the burden of rejoinder to the State. The rec*1093ord provides no information as to rural white voters. The district court viewed that as fatal to the plaintiffs’ claims. In fact, it meant that the plaintiffs had satisfied their threshold requirements, and the State had failed to rejoin. The district court erred in holding the plaintiffs to a higher evidentia-ry burden.
C
The district court did not reach the second prong .of the § 2 analysis, namely, whether the burden was in part caused by or linked to social and historical conditions that have or currently produce discrimination against members of the protected class. Nevertheless, the plaintiffs established a likelihood of success on the second prong.
As to the second part of the analysis, the Supreme Court has identified several factors to be taken into consideration, consistent with the legislative history of the Voting Rights Act, namely:
(1) the extent of any history of official discrimination in the state or political subdivision that touched the right of the members of the minority group to register, to vote, or otherwise to participate in the democratic process;
(2) the extent to which voting in the elections of the state or political subdivision is racially polarized;
(3) the extent to which the state or political subdivision has used unusually large election districts, majority vote requirements, anti-single shot provisions, or other voting practices or procedures that may enhance the opportunity for discrimination against the minority group;
(4) if there is a candidate slating process, whether the members of the minority group have been denied access'to that process;
(5) the extent to which members of the minority group in the state or political subdivision bear the effects of discrimination in such areas as education, employment and health, which hinder then-ability to participate effectively in the political process;
(6) whether political campaigns have been characterized by overt or subtle racial appeals; and
(7) the extent to which members of the minority group have been elected to public office in the jurisdiction.
Gingles, 478 U.S. at 37, 106 S.Ct. 2752. In addition, the Court added that in some cases, there was probative value in inquiring “whether there is a significant lack of responsiveness on the part of elected officials to the particularized needs of the members of the minority group” and “whether the policy underlying the state or political subdivision’s use of such voting qualification, prerequisite to voting, or standard, practice or procedure is tenuous.” Id. (citing S. Rep., at 28-29, U.S.Code Cong. & Admin. News 1982, pp. 206-207).
As to the first factor, the extent of any history of official discrimination in the state or political subdivision that touched the right of the members of the minority group to register, to vote, or otherwise to participate in the democratic process, Arizona has had a long history of imposing burdens on minority voters. In 1912, shortly after gaining statehood, Arizona imposed a literacy test for voting. In Cochise and Pima Counties, the denial of the right to vote meant that nearly half the precincts lacked enough voters to justify holding primary elections in 1912. From 1912 to the early 1960s, election registrars applied the literacy test to reduce the ability of African Americans, Native Americans, and Hispanics to register to vote. In an action filed against Arizona to enforce the Voting Rights Act, the United States Justice Department estimated that 73,000 peo-*1094pie could not vote because of the existence of the literacy test.
The passage of the Voting Rights Act in 1965 caused the suspension of the literacy test in Arizona, but the statute remained in effect until it was repealed in 1972, after Congress banned its use in 1970 through an amendment to the Voting Rights Act. Arizona subsequently unsuccessfully challenged the Congressional ban on literacy tests. Oregon v. Mitchell, 400 U.S. 112, 118, 91 S.Ct. 260, 27 L.Ed.2d 272 (1970). In Mitchell, the Court noted that, in Arizona, only two counties out of eight with Hispanic populations in excess of 15% showed voter registration equal to the state-wide average. Id. at 132, 91 S.Ct. 260. In the 1960s, there were a number of initiatives to discourage minority voting in Arizona, such as “Operation Eagle Eye.” Under Operation Eagle Eye, minority voters were challenged at the pools on á variety of pretexts, with the goal of preventing minority voting or slowing down the process to create long lines at the polls and discourage voting.
Native Americans in Arizona especially suffered from voting restrictions. Although Native Americans were U.S. citizens, the Arizona Supreme Court held in 1928 that they could not vote because they were under federal guardianship. Porter v. Hall, 34 Ariz. 308, 271 P. 411, 419 (1928). Even after that ban was overruled in 1948 in Harrison v. Laveen, 67 Ariz. 337, 196 P.2d 456 (1948), Native Americans faced significant obstacles to voting. See generally, Patty Ferguson-Bohnee, The History of Indian Voting Bights in Arizona: Overcoming Decades of Voter Suppression, 47 Ariz. St. L. J. 1099, 1112 (2015).
Because of its long history of imposing burdens on minority voting, Arizona became one of nine states subject to the pre-clearance requirements of the Voting Rights Act after it was amended in 1975 to protect language minorities. 40 Fed. Reg. 43746. Under the pre-clearance provision, Arizona was required to obtain the approval of the United States Department of Justice ■ before implementing any law affecting the voting rights and representations of minorities. Since 1982, the Department of Justice has vetoed four statewide redistricting plans proposed by Arizona that appeared to discriminate against minorities. When Arizona was subject to the pre-clearance requirements of § 5, a bill precluding or criminalizing ballot collection passed the Arizona legislature, but was ultimately repealed due to concerns about Justice Department approval. In 2013, the Arizona legislature passed a measure banning partisan ballot collection, the violation of which was a misdemeanor. It was repealed after its repeal was placed on the ballot by referendum. The plaintiffs established a likelihood of success as to the first factor.
As to the second factor, the extent to which voting in the elections of the state or political subdivision is racially polarized, Arizona has had a history óf racially polarized voting. The plaintiffs provided expert testimony detailing the history of polarized voting. Statistical analysis showed the sharp polarization between white and nonwhite voters.
For the reasons described in the discussion of factor one, the plaintiffs demonstrated a likelihood of success as to factor three, namely, the extent to which the state or political subdivision has used unusually large election districts, majority vote requirements, anti-single shot provisions, or other voting practices or procedures that may enhance the opportunity for discrimination against the. minority group.
Because the.-voting access issues affect the right to vote for a candidate,- the fourth factor concerning the candidate slating process is not relevant.
*1095The fifth factor, the extent to which members of the minority group in the state or political subdivision bear the effects of discrimination in such areas as education, employment, and health, which hinder their ability to. participate effectively in. the political process, falls decisively in favor of the plaintiffs. The plaintiffs tendered significant evidence showing that Arizona minorities suffered in education and employment opportunities, with disparate poverty rates, depressed wages, higher levels of. unemployment, lower educational attainment, less access to transportation, residential transiency, and poorer health.
The plaintiffs also provided substantial evidence as to the sixth factor, namely, whether political campaigns have been characterized by overt or subtle racial appeals.
Finally, the plaintiffs provided evidence supporting the seventh Gingles factor, namely, the extent to which members of the minority group have been elected to public office in the jurisdiction. As of January 2016, Hispanics constituted over 30% of the population, but held only 19% of the seats in the Arizona legislature. African-Americans made up 4.7% of the population, but held 1% of the legislative seats. Native Americans fared slightly better, constituting 5.3% of the population and holding 4.4% of the legislative seats.
. But the Gingles factors are not the end of the story. We are obligated to look to the “totality of the circumstances.” 52 U.S.C. § 10301(b). In this election, in-person voting opportunities are significantly hindered by lack of polling places and significant changes in polling places, all of which have caused extraordinarily long lines for voting in person, up to six hours in some locations. This hindrance to in-person voting falls most heavily on minorities, So, the cited “opportunities” for .alternate voting are illusory. H.B. 2023 has now imposed additional significant burdens on minorities as to their ability to cast their ballots early through the popular means of ballot collection. The totality of the circumstances of this election, coupled with the historic discrimination in Arizona’s electoral politics are sufficient to satisfy the second § 2 requirement. In sum, the plaintiffs established a likelihood’ of success in proving the Gingles factors at stage two of the § 2 analysis.
D
The plaintiffs established a likelihood of success on - the § 2 Voting Rights Act claim. They established that the criminalization of ballot collection meant that minority voters had less opportunity than other members of the electorate to elect representatives of their choice, and that the burden in part was caused by or linked to social and historical conditions that have or currently produce discrimination against minorities,
IV
The district court should have granted the motion for a preliminary injunction. The district court made a number of legal errors. The plaintiffs established that the anti-ballot-collection law significantly burdens the voting rights of minorities, particularly Hispanic and Native American voters. The State’s justification of preventing voter fraud .was not, and -is not, supportable. One of the most popular and effective methods of minority voting is now a crime. H.B. 2023 violates the Constitution and the Voting Rights Act.
There are many burdens and challenges faced in Arizona by Native Americans, Hispanics, African-Americans, the poor, and the infirm who do not have caregivers *1096or family. With H.B. 2023, Arizona has added another: disenfranchisement.
I respectfully dissent.
. The majority believes the district court’s findings of fact are reviewed by this Court for clear error because the district court has superior fact-finding capabilities. Maj. Op. at 1071. The majority also believes a district court’s answer to the ultimate question— whether there was a § 2 violation—-is a finding of fact entitled to deference. The majority cites Gonzales for that proposition. However, the district court did not conduct any eviden-tiary hearings to resolve disputed factual issues, and most of the record is- undisputed, and the parties’ submissions were by affidavit. Furthermore, the district court here did not determine whether there was a § 2 violation because, unlike in Gonzales, we are not yet at the merits stage of the inquiry. This is an appeal of a denial of a preliminary injunction, so we are reviewing the district court’s determination that the plaintiffs are unlikely to succeed on the merits of their claims: In my view, the plaintiffs are likely to succeed on the merits and the district court reached the opposite conclusion because it made errors of law. Therefore, review is de novo as to those questions. Pom Wonderful LLC, 775 F.3d at 1123. Most of the district court’s opinion involves a mixed question of law and fact. In election cases, as with other appeals, we review such decisions de novo. United States v. Blaine County, Montana, 363 F.3d 897, 909 (9th Cir. 2004).
. The majority concludes that because Arizona’s regulatory interests are sufficient to justify the "minimal burden” imposed by H.B. 2023, "the district court was not required to conduct a' means-end fit analysis here.” Maj. Op', at 1082. That is an erroneous interpretation of Supreme Court and our precedent. “The Supreme Court delineated the appropriate standard of review for laws regulating the right to vote in Burdick v. Takushi[:]" it is a "balancing and means-end fit framework.” Pub. Integrity All., 836 F.3d at 1024. A court may not avoid application of a means-end fit framework in favor of rational basis. review simply by concluding that the state’s regulatory interests justify the voting burden imposed. Moreover, Burdick tells us that in weighing "the character and magni*1086tude of the asserted injury” against the "precise interests put forward by the State as justifications for the burden imposed by its rule,” we must take into consideration -“the extent to which those interests make it necessary to burden the plaintiff’s rights.” 504 U.S. at 434, 11-2 S.Ct. 2059. In this case, the State’s asserted interest does not make necessary the substantial burden on the voting rights of minorities. Simply put, the State’s end does not fit the means employed.
. The majority asserts that plaintiffs in this case are bringing a facial challenge to H.B. 2023 and they therefore bear a “heavy burden of persuasion” because such challenges "raise the risk of premature interpretation of statutes.” Maj. Op. at 1079 (internal quotations omitted). It is worth noting that neither the plaintiffs nor the defendants categorize the challenge to H.B. 2023 as a facial challenge; only the majority opinion does so. It is also worth noting that securing a court’s interpretation of the effects of H.B. 2023 before the law is enforced is the point of seeking a preliminary injunction. But for my part, I think this is a distinction without a difference because "[t]he underlying constitutional standard [in an as applied challenge] ... is no different th[a]n in a facial challenge.” Legal Aid Servs. of Or. v. Legal Servs. Corp., 608 F.3d 1084, 1096 (9th Cir, 2010) (quoting Velazquez v. Legal Servs. Corp., 462 F.3d 219, 228 (2d Cir. 2006)). “Facial and as-applied challenges differ in the extent to which the invalidity of a statute need be demonstrated (facial, in all applications; as-applied, in a personal application). Invariant, however, is the substantive rule of law to be used. In other words, how one must demonstrate the statute’s invalidity remains the same for both types of challenges, namely, by showing that a specific rule of law, usually a constitutional rule of law, invalidates the statute, whether in a personal application or to all.” Velazquez, 462 F.3d at 228 (emphasis in original).
. Plaintiffs assert an additional Constitutional claim under the First Amendment. In my view, the district court erred in concluding that H.B. 2023 did not burden their First Amendment associational rights. However, in my view, the district court did not abuse its discretion in denying a preliminary injunction based on this independent claim.
. The majority opines that "[w]hile § 2 itself does not require quantitative evidence, past cases suggest that such evidence is typically necessary to establish a disparate impact.” Maj. Op. at 1073. The majority also notes that plaintiffs' briefs rely on vote dilution cases but not vote denial cases in arguing that statistical evidence is not required to establish a § 2 violation. Maj. Op. at 1073 fn. 14. I perceive no reason why the type of § 2 case on which plaintiffs rely is of consequence to their argument about what § 2 itself requires. Likely plaintiffs could not rely on a vote denial case for the stated proposition because of the practical reality that in a vote denial case, quantitative evidence of the effect of a rule on voting behavior is only available after an election has occurred, at which point the remedial purpose of the Voting Rights Act is no longer served. Plaintiffs in vote dilution cases, in contrast, can often gather and analyze quantitative data before an election. See, e.g., Thornburg v. Gingles, 478 U.S. 30, 106 S.Ct. 2752, 92 L.Ed.2d 25 (1986).