Defendants, State of Washington officials, appeal the district court’s decision that a Washington statute prohibiting exit polling within 300 feet of the polling place violated the First and Fourteenth Amendments as applied to plaintiffs, three television stations (American Broadcasting Companies, Inc., CBS Inc., and the National Broadcasting Company, Inc.) and two newspapers (The Daily Herald Company and The New York Times Company) (“media plaintiffs”). We affirm the district court.
I.
This case is before this court for the second time. See Daily Herald Co. v. Munro, 758 F.2d 350 (9th Cir.1985) (“Daily Herald /”). It involves Wash.Rev.Code § 29.51.020, a statute regulating conduct in and around polling places on election day. Violating the statute is a misdemeanor. Before 1983, the statute prohibited certain activities within 100 feet of the polling place. In 1983, the Washington Legislature amended the statute by adding subsections (l)(d) and (l)(e) and extending the area to 300 feet:
(1) On the day of any primary, general or special election, no person may, within a polling place, or in any public area within three hundred feet of any entrance to such polling place:
(d) Engage in any practice which interferes with the freedom of voters to exercise their franchise or disrupts the administration of the polling place; or
(e) Conduct any exit poll or public opinion poll with voters.
1983 Wash.Laws 1st Ex.Sess., ch. 33, § 1.
Although the statute does not define the term “exit poll,” at oral argument the parties agreed that an exit poll was a systematic questioning of voters as they left the polling place. The media plaintiffs’ systematic requests of voters leaving the polling place that they volunteer to complete written questionnaires constituted exit polling. The state agreed that random ques*383tioning or interviewing of voters by journalists is not exit polling and is not prohibited. In making this distinction, the state explained that the legislature considered exit polling, but not random interviewing, to be a problem.
On December 12, 1983, the media plaintiffs filed this 42 U.S.C. § 1983 action in federal district court against two Washington state officials in their official capacities to enjoin enforcement of Wash.Rev.Code § 29.51.020(1 )(e), the exit polling statute, alleging that enforcement violated the First and Fourteenth Amendments. The defendants counterclaimed under section 1983 and the Voting Rights Act, 42 U.S.C. § 1971, for an injunction to prevent the media plaintiffs from violating the statute, alleging that their conduct infringed on Washington state voters’ rights by disrupting the peace, order, and decorum of the polling place, and discouraging voting.
The district court found the statute constitutional in a summary judgment proceeding, and a panel of this court reversed and remanded for trial. Daily Herald I, 758 F.2d at 351. On remand, the district court heard evidence to answer eight factual questions posed by the panel. See id. at 351-52. The district court then held the statute unconstitutional as applied to the plaintiffs, and ruled against the state on its counterclaim. The district court found that the media plaintiffs conducted their exit polling in a “systematic and statistically reliable manner”; that information obtained from exit polling could not be obtained by other methods; that the 300-foot limit precluded exit polling; and that exit polling was not per se disruptive to the polling place.1
On appeal, the state argues that the district court erred in holding that the statute was not the least restrictive means of advancing the state’s interests, and in refusing to allow some of its witnesses to testify. The media plaintiffs argue that the statute unconstitutionally infringes on their First Amendment rights to gather and broadcast news.
II.
We review de novo the “district court’s application of the law to the facts on free speech questions.” Jews for Jesus, Inc. v. Board of Airport Commissioners, 785 F.2d 791, 792 (9th Cir.1986), aff'd, — U.S. -, 107 S.Ct. 2568, 96 L.Ed.2d 500 (1987). The Constitution requires different standards for reviewing findings of fact in those cases in which the district court upholds a restriction on speech as constitutional than in those in which it does not. See Planned Parenthood Association/Chicago Area v. Chicago Transit Authority, 767 F.2d 1225, 1228-29 (7th Cir.1985) (citing Bose Corp. v. Consumers Union of United States, Inc., 466 U.S. 485, 104 S.Ct. 1949, 80 L.Ed.2d 502 (1984), and New York Times Co. v. Sullivan, 376 U.S. 254, 84 S.Ct. 710, 11 L.Ed.2d 686 (1964)). When a district court holds a restriction on speech constitutional, we conduct an independent, de novo examination of the facts. Id. When the government challenges the district court’s holding that the government has unconstitutionally restricted speech, on the other hand, we review the district court findings of fact for clear error. Id. This rule “reflects a special solicitude for claims that the protections afforded by the First Amendment have been unduly abridged,” id. at 1229, while not affording special protection “for the government’s claim that it has been wrongly prevented from restricting speech,” id. In this case, therefore, we review the district court’s findings of fact for clear error.
III.
In First Amendment cases, when a statute “covers the particular conduct” of the statute’s challengers, it is “proper to reach the constitutional question involved in th[e] case.” United States v. Grace, 461 *384U.S. 171, 176, 103 S.Ct. 1702, 1706, 75 L.Ed. 2d 736 (1983). The media plaintiffs’ conduct violates the statute; we therefore reach the constitutional question. However, we address only subsection (l)(e), because that subsection is severable from the rest of the statute.2 See Carey v. Brown, 447 U.S. 455, 459 n. 2, 100 S.Ct. 2286, 2289 n. 2, 65 L.Ed.2d 263 (1980).
The First Amendment provides that “Congress shall make no law ... abridging the freedom of speech.” The media plaintiffs’ exit polling constitutes speech protected by the First Amendment, not only in that the information disseminated based on the polls is speech, but also in that the process of obtaining the information requires a discussion between pollster and voter. “ ‘[A] major purpose of [the First] Amendment was to protect the free discussion of governmental affairs. This of course includes discussions of candidates ____’” Brown v. Hartlage, 456 U.S. 45, 52-53, 102 S.Ct. 1523, 1528, 71 L.Ed.2d 732 (1982) (quoting Mills v. Alabama, 384 U.S. 214, 218, 86 S.Ct. 1434, 1436, 16 L.Ed. 2d 484 (1966)). Moreover, the First Amendment protects the media’s right to gather news. See, e.g., Richmond Newspapers, Inc. v. Virginia, 448 U.S. 555, 576, 100 S.Ct. 2814, 2827, 65 L.Ed.2d 973 (1980) (“Free speech carries with it some freedom to listen.”); Branzburg v. Hayes, 408 U.S. 665, 681, 92 S.Ct. 2646, 2656, 33 L.Ed.2d 626 (1972) (“[W]ithout some protection for seeking out the news, freedom of the press could be eviscerated.”); see also In re Express News Corp., 695 F.2d 807 (5th Cir.1982); United States v. Sherman, 581 F.2d 1358 (9th Cir.1978).3 Exit polling is thus speech that is protected, on several levels, by the First Amendment.4
The state may regulate exit polling only in limited ways. The standard we use to determine whether the regulation is constitutional depends on whether the speech to be regulated occurs in a traditional public forum, see Cornelius v. NAACP Legal Defense & Educational Fund, Inc., 473 U.S. 788, 797, 105 S.Ct. 3439, 3446, 87 L.Ed. 2d 567 (1985), and whether the statute is content-based or amounts to an absolute prohibition of a type of speech, see United States v. Grace, 461 U.S. 171, 177, 103 S.Ct. 1702, 1706, 75 L.Ed.2d 736 (1983).
The public areas within 300 feet of the entrance to the polling place are traditional public forums because they traditionally are open to the public for expressive purposes, including random interviews by reporters, and encompass streets and sidewalks.5 See Perry Education Association v. Perry Local Educators’ Association, 460 U.S. 37, 45, 103 S.Ct. 948, 954, 74 *385L.Ed.2d 794 (1983). The statute is content-based because it regulates a specific subject matter, the discussion of voting, see Consolidated Edison Co. of New York v. Public Service Commission, 447 U.S. 530, 537, 100 S.Ct. 2326, 2333, 65 L.Ed.2d 319 (1980), and a certain category of speakers, exit pollsters, see Widmar v. Vincent, 454 U.S. 263, 267-70, 102 S.Ct. 269, 273-74, 70 L.Ed.2d 440 (1981).
A content-based statute that regulates speech in a public forum is constitutional only if it is narrowly tailored to accomplish a compelling government interest, Consolidated Edison Co., 447 U.S. at 540, 100 S.Ct. at 2334, and is the least restrictive means available, Cornelius, 473 U.S. at 800, 105 S.Ct. at 3448; Cinevision Corp. v. City of Burbank, 745 F.2d 560, 569 (9th Cir.1984), cert. denied, 471 U.S. 1054, 105 S.Ct. 2115, 85 L.Ed.2d 480 (1985). “[Regulations enacted for the purpose of restraining speech on the basis of its content presumptively violate the First Amendment.” City of Renton v. Playtime Theatres, Inc., 475 U.S. 41, 46-47, 106 S.Ct. 925, 928, 89 L.Ed.2d 29 (1986) (citing Carey v. Brown, 447 U.S. at 462-63 & n. 7, 100 S.Ct. at 2291 & n. 7, and Police Department of Chicago v. Mosley, 408 U.S. 92, 95, 98-99, 92 S.Ct. 2286, 2289, 2291-92, 33 L.Ed.2d 212 (1972)).6
The Washington statute is not narrowly tailored to advance the state’s interest. States have an interest in maintaining peace, order, and decorum at the polls and “preserving the integrity of their electoral processes.” Brown v. Hartlage, 456 U.S. 45, 52, 102 S.Ct. 1523, 1528, 71 L.Ed.2d 732 (1982); Mills v. Alabama, 384 U.S. 214, 218, 86 S.Ct. 1434, 1436, 16 L.Ed.2d 484 (1966); see also Note, Exit Polls and the First Amendment, 98 Harv.L.Rev. 1927, 1933 (1985).7 However, the statute is not narrowly tailored to advance this interest. The state argues that the statute’s purpose is to prevent disruption at the polling place, but the statute prohibits all exit polling, including nondisruptive exit polling. Prohibiting nondisruptive exit polling therefore does not advance the state’s interest, and also renders the statute overbroad because it applies to activities not implicating the interest.
Moreover, the statute is not the least restrictive means of advancing the state’s interest. The statute is unnecessarily restrictive because Wash.Rev.Code § 29.51.020(l)(d) already prohibits disruptive conduct at the polls. Daily Herald-I, 758 F.2d at 359 (Norris, J., concurring in part and dissenting in part). We also note that several other less restrictive means of advancing this interest exist: for example, reducing the size of the restricted area; requiring the media to explain that the exit poll is completely voluntary; requiring polling places to have separate entrances and exits, see Note, Exit Polls and the First Amendment, 98 Harv.L.Rev. 1927, 1935 (1985); or prohibiting everyone except election officials and voters from entering the polling room. Of course, we do not suggest that a statute using one or all of these less restrictive means would necessarily pass constitutional muster; we list these alternatives simply to demonstrate the deficiency of the statute as currently written.8
*386We conclude that the statute is unconstitutional on its face. It is content-based, overbroad, and not the least restrictive means of advancing the state's legitimate interest of keeping peace, order, and decorum at the polls.
IV.
Even if the statute were deemed to be content-neutral, it would still be facially invalid. The statute is not a reasonable time, place, and manner regulation, narrowly tailored to serve a significant government interest and leaving open ample alternative channels of communication. Consolidated Edison Co., 447 U.S. at 535-36, 100 S.Ct. at 2332 (citations omitted); see also Renton, 475 U.S. at 47, 106 S.Ct. at 929.9 We hold that the district court did not clearly err in finding that the restriction amounts to a prohibition on all exit polling and that alternative polling techniques would not produce the same information, or constitute as scientific a study. The district court could have concluded from the evidence that, for example, persons contacted by telephone would be less likely to be home, or less willing to participate. Moreover, the media plaintiffs argue, a voter contacted by telephone would less likely be candid about whether and how he or she voted because of exposure to election results, editorials, other persons’ opinions, or lack of anonymity. Therefore, no alternative channels of communication exist to gather the type of information obtained through exit polling.
The state argues that the media plaintiffs could obtain the same information by telephone because exit polling is not “scientific”: deviations from the practice of selecting voluntary voters on a predetermined basis destroy the gathering of a scientific sample. For example, some voters volunteered for the poll even though they were not part of the predetermined sample, and some were not “voluntary” because they thought the poll was required. Some deviation from the process may render it less scientific, but studies are usually adjusted to account for deviations, and the district court was not clearly erroneous in disregarding these deviations.
In addition, the existence of an alternative may not justify a regulation affecting speech; it is only “a factor to be considered in striking the appropriate accommodation between free speech and legitimate governmental interests.” Baldwin v. Redwood City, 540 F.2d 1360, 1368 (9th Cir.1976), cert. denied, 431 U.S. 913, 97 S.Ct. 2173, 53 L.Ed.2d 223 (1977).
Thus, even if the statute were somehow content-neutral, we believe the restriction is not a reasonable time, place, or manner regulation. For the reasons discussed in the previous section, it is not narrowly tailored to achieve a significant government interest. Moreover, it leaves open no alternative channels of discussing or obtaining information uniquely derived from exit polling.
V.
The media plaintiffs argue that the statute is unconstitutional for another reason: that the stated purpose for the statute of protecting order at the polls was a pretext, and that the state’s true motive was to prevent the media from broadcasting election results before the polls closed. The media plaintiffs argue that this pur*387pose is insufficient to justify speech regulation because it amounts to a general interest in insulating voters from information on election day. Mills v. Alabama, 384 U.S. 214, 86 S.Ct. 1434, 16 L.Ed.2d 484 (1966). The district court orally found that a purpose of the statute was to prevent broadcasting of early returns. The state argues that this purpose was not a motive in passing the statute but that, if it were, such a purpose would be legitimate and the statute would be constitutional.
We hold that, assuming that at least one purpose of the statute was to prevent broadcasting early returns, the statute is unconstitutional because this purpose is impermissible and because the statute is neither narrowly tailored to advance the state’s interest nor the least restrictive alternative.10
The state argues that, although this purpose was not a justification for the statute, such broadcasting does affect voters’ decisions about whether to vote, and the state has a significant interest in preventing this influence on voting behavior. Some studies indicate that the broadcasting of early predictions in 1980, together with President Carter’s early concession speech, may have affected voter behavior. Note, Exit Polls and the First Amendment, 98 Harv.L. Rev. 1927, 1929-30 & nn. 16-26 (1985). However, these studies showed that it was the early projection of the results of polls from eastern states that may have affected voters in the western states, so the statute does not advance this interest at all. See id. at 1940. The state offers no evidence to the contrary.
Furthermore, a general interest in insulating voters from outside influences is insufficient to justify speech regulation. Just as with election-day broadcasts or newspaper editorials that may affect voters’ choices, regulating the speech here on the basis that it might indirectly affect the voters’ choice is impermissible. See Mills v. Alabama, 384 U.S. 214, 218-20, 86 S.Ct. 1434, 1436-37, 16 L.Ed.2d 484 (1966); Vanasco v. Schwartz, 401 F.Supp. 87, 100 (S.D.N.Y.1975) (three-judge court) (“[W]hen the State through the guise of protecting the citizen’s right to a fair and honest election tampers with what it will permit the citizen to see and hear even that important state interest must give way to the irresistible force of protected expression under the First Amendment.”), aff'd mem., 423 U.S. 1041, 96 S.Ct. 763, 46 L.Ed.2d 630 (1976).
In addition, the statute is not narrowly tailored to protect voters from the broadcasting of early returns.11 Election-day *388broadcasting is only one use to which the media plaintiffs put the information gathered from exit polling and the statute, but by prohibiting exit polling the statute prohibits speech involving other uses of the information. The New York Times, for example, has used information obtained from exit polls to report, among other things, that the 1980 election was essentially “a rejection of President Carter more than a conservative revolution.” The television networks have used the information to analyze the results of elections on election night and in subsequent news programming, and to distribute their polling results to third parties after each election, including scholars who study and report on the American electoral process generally. The effect of the statute on these forms of speech, none of which involve protecting voters, render the statute unconstitutional.
Thus, the statute is unconstitutional as a protection for Washington voters because it fails each part of the test: it is not narrowly tailored to advance a significant government interest, and not the least restrictive means available. Because the effect of the statute is to prohibit more speech than broadcasting of early returns, which is the evil the state seeks to prevent, it is not narrowly tailored. Because the general interest of insulating voters from outside influences is not sufficient to justify speech regulation, it is not a significant state interest. Because no evidence suggests that the broadcast of early returns from the state of Washington affects Washington voters’ behavior, it does not advance the state’s interest. And, for the reasons discussed in the previous section, it is not the least restrictive means available.
VI.
Finally, the state argues that the district court abused its discretion in refusing to allow three witnesses, two of them experts, to testify at trial. We review decisions regarding admission of evidence for abuse of discretion. Brocklesby v. United States, 767 F.2d 1288, 1292 n. 1 (9th Cir.1985), cert. denied, 474 U.S. 1101, 106 S.Ct. 882, 88 L.Ed.2d 918 (1986). In addition, the district court has great latitude in deciding whether to allow expert witnesses to testify, and we uphold the court’s action unless it is manifestly erroneous. United States v. Gann, 732 F.2d 714, 724 (9th Cir.), cert. denied, 469 U.S. 1034, 105 S.Ct. 505, 83 L.Ed.2d 397 (1984). The discussion between the court and the attorneys about the state’s offers of proof for witnesses testifying for the state show that the court did not abuse its discretion in excluding their testimony.
First, the state made an offer of proof that Dan Montgomery, the sheriff in Thurston County, was asked a series of questions in the polling place by an exit pollster, thought this pollster was a state official, and felt “accosted.” However, the state conceded that this testimony was cumulative. In addition, the media plaintiffs challenge enforcement of the statute around, not in, the polling place, so the testimony was not relevant.
Second, the state made an offer of proof that Jack Bryar, the Secretary of State of Kansas, would testify as an expert witness about the effect of exit polling on polling places in Kansas. The court decided that it was aware of anything relevant about which Bryar would testify, and that the testimony would be cumulative of defendant Munro, who also testified as an expert.
Third, the state made an offer of proof that Curtis Gange, as an expert on elections, would testify about the decline of voting in Washington, a study of detrimental effects of broadcasting the information obtained through exit polls before the polls close. The district court found the proffered testimony irrelevant because the state had presented no evidence that any voter in Washington declined to vote because of broadcasting of early returns. In addition, the state conceded that some of his testimony would be cumulative.
When the district court was satisfied after extensive colloquy with counsel for the state that the proffered testimony was cumulative or irrelevant, it exercised its discretion to refuse the testimony. We hold *389that the district court did not abuse its discretion.
VII.
Wash.Rev.Code § 29.51.020(l)(e), which prohibits exit polling within 300 feet of the polling place on election day, is unconstitutional on its face. The statute regulates speech on the basis of content in a traditional public forum. It is not narrowly tailored to advance a significant government interest, whether that interest is keeping peace, order, and decorum at the polls, or insulating voters from the influence of early election return projections, and it is not the least restrictive means available.
AFFIRMED.
. In ruling against the defendants on their counterclaim, the district court found that the media plaintiffs could not be liable under section 1983 because their conduct did not constitute state action, and that they could not be liable under section 1971 because they lacked intent to interfere with voting. The defendants have not appealed the order denying relief on the counterclaim for an injunction.
. The state argues that it is possible to sever and invalidate the 300-foot provision, allowing the former 100-foot radius to come back into effect. We disagree. The Washington Legislature amended the statute, and invalidation of a later version of a statute does not resurrect the former version.
. The state argues this case is a “right to access” case, and that the speech involved here would not be entirely protected under the right to access cases. The right to access is discussed in the separate concurring opinion of Judge Reinhardt. We do not decide the case on the basis of that principle. However, we do not agree with the state that the conduct here would not be entitled to protection under the "right to access” cases, and note that even where the right to access is qualified, any restriction must be narrowly tailored to serve a compelling governmental interest. Press-Enterprise Co. v. Superior Court, 464 U.S. 501, 510, 104 S.Ct. 819, 824, 78 L.Ed.2d 629 (1984); Globe Newspaper Co. v. Superior Court, 457 U.S. 596, 607, 102 S.Ct. 2613, 2620, 73 L.Ed.2d 248 (1982).
. The state also argues that exit polling is closer to commercial speech than political speech, and therefore less deserving of constitutional protection. A profit motive by the media plaintiffs is irrelevant to the inquiry of whether the content of their speech is political or commercial. Pittsburgh Press Co. v. Pittsburgh Commission on Human Relations, 413 U.S. 376, 385, 93 S.Ct. 2553, 2558, 37 L.Ed.2d 669 (1973). Newsgather-ing is not commercial speech; commercial speech "‘does "no more than propose a commercial transaction.” ’ ” Bolger v. Youngs Drug Products Corp., 463 U.S. 60, 66, 103 S.Ct. 2875, 2880, 77 L.Ed.2d 469 (1983) (quoting Virginia State Board of Pharmacy v. Virginia Citizens Consumer Council, 425 U.S. 748 at 762, 96 S.Ct. 1817 at 1825, 48 L.Ed.2d 346 (1976).
. The media plaintiffs concede that the state may constitutionally limit activity inside the polling place to voting, and do not challenge the state's enforcement of the statute within the polling place itself.
. Statutes that are content-based because they regulate subject matter are susceptible to analysis under the Equal Protection Clause as well as under the First Amendment. Carey v. Brown, 447 U.S. at 459, 461-62, 100 S.Ct. at 2289, 2290-91. The standard for scrutinizing content-based statutes is the same under either amendment.
. The media plaintiffs assert that, in passing the statute, the state intended to prevent the broadcasting of election-night projections. We discuss this contention in Part V.
. The state argues that exit polling interferes with Washington state citizens’ right to vote. However, the district court found as a fact that the media’s conduct was not shown to be disruptive or intended to interfere with any voter’s right to vote. That finding is not clearly erroneous; as the district court pointed out during the closing arguments, there was no evidence that someone did not vote or voted differently because of the exit polls. Reporter’s Transcript, Vol. Ill at 168, lines 11-13 ("[Tjhere is no testimony or evidence that one single voter in Washington didn’t vote because of exit polling____”); Reporter’s Transcript, Vol. Ill at 170, lines 20-22 ("There isn’t one iota of testimony about a single voter that was upset, or intimidated, or threatened.”).
The state also argued that, although it did not present testimony at trial showing specific in*386stances of voters who were discouraged from voting, the legislative history of the statute contains such evidence. After reviewing the legislative hearing for this bill, however, we find no such evidence. The hearing describes one instance of an exit pollster blocking voters’ entrance to the polling booths, but the media plaintiffs agree the state can regulate such activity in the polling place. This instance does not, therefore, help us decide the issue before us.
At oral argument, the state argued that disruption meant not only physical disruption, but also a violation of the decorum of the voting area. However, we cannot say that peaceful, systematic passing out of questionnaires has a detrimental effect on decorum as a matter of law.
. The statute is unreasonable because, as we explained in the previous section, it prohibits even nondisruptive exit polling. Also, to the extent that a time, place, and manner regulation must be the least restrictive means available, Cinevision Corp., 745 F.2d at 569, the least restrictive means analysis of the previous section applies.
. The state argues that we cannot consider this justification as a possible purpose of the statute. The state argues that, although some state officials, including defendant Munro, testified before the Washington Legislature about the detrimental effect of broadcasting early returns, the legislative history of the statute does not establish that reason as a justification for enacting the statute.
Although this argument raises some interesting questions about what constitutes the relevant legislative history, this motive was discussed at length during the legislative hearings and appears to have been at least one purpose for the statute. In addition, the media plaintiffs introduced into evidence letters from Munro to news media organizations and Munro’s testimony before Congress and the Washington Legislature. These show some concerns about the effect on voting of early projections based on exit polling information. Two law review notes that discuss this case assume this is the state legislature’s true purpose, and cite to Washington state officials’ testimony before Congress when Congress was deciding whether to pass a statute limiting broadcasting early returns. See Note, Exit Polls and the First Amendment, 98 Harv.L. Rev. 1927, 1932, 1938-39 (1985); Note, Curtailment of Early Election Predictions: Can We Predict the Outcome?, 36 U.FlaJL.Rev. 489 (1984). Moreover, the district court did not make a finding that preserving order was the sole purpose of the statute.
In analyzing the statute applying each of the asserted purposes, we need not decide which motive was the true purpose; we find the statute unconstitutional on its face whether the purpose was the first, the second, or both.
. During the legislative hearings for this amendment, the Washington Legislature questioned whether a state could enact legislation that regulated broadcasting. Of course, this question raises serious issues involving the exclusiveness of the federal government’s authority to regulate broadcasting, and the validity of an otherwise constitutional statute with a secondary effect of regulating broadcasting. However, we need not resolve these issues because we find the statute unconstitutional under more traditional First Amendment tests.