ORDER
The opinion and dissenting opinion, filed June 25, 2002, slip op. at 9009, 292 F.3d 1192, are withdrawn and replaced by the amended opinion, concurring opinion, and dissenting opinion filed concurrently with this order. With these amendments, the panel has voted to deny the petition for rehearing en banc, with Judge Ferguson recommending that the petition be granted.
The full court was advised of the petition for rehearing en banc. A judge of the court called for a vote on whether to rehear the matter en banc. On such vote, a majority of the nonrecused active judges failed to vote in favor of en banc rehearing.*
The petition for rehearing en banc is denied.
KOZINSKI, Circuit Judge, with whom PREGERSON, REINHARDT, T.G. NELSON, MICHAEL DALY HAWKINS, THOMAS, McKEOWN, WARDLAW, WILLIAM A. FLETCHER, FISHER and BERZON, Circuit Judges, join,dissenting from denial of rehearing en banc.
For over half a century, Consumers Union has been testing and rating consumer products and publishing the results in its magazine, Consumer Reports. A significant portion of the American public relies on CU’s ratings on a regular basis, and almost everyone consults Consumer Reports now and then before making a significant purchase — whether a sound system, a dishwasher or a car. What makes CU’s ratings particularly useful is the thorough explanation of the testing procedures employed, which lets consumers judge whether the ratings fairly represent the product.
The Suzuki Samurai article, the subject of this lawsuit, is no exception. Running some 6500 words, it tells readers precisely how CU came to conclude that “The Suzuki rolls over too easily,” starting with an incident during the vehicle’s break-in period where the Samurai “flopped over on its side” during a low-speed maneuver. The explanation is not written for morons; like other CU reviews, it is geared to an intelligent, informed consumer. Yet the careful reader will not fail to understand the central facts that undergird Suzuki’s claim in this lawsuit, namely, that the Samurai did well on CU’s standard course, that CU then modified the course to make it more challenging and, as a result, the Samurai did far worse than its competitors.
I find it incomprehensible that a review truthfully disclosing all this information could be deemed malicious under New York Times Co. v. Sullivan, 376 U.S. 254, 84 S.Ct. 710, 11 L.Ed.2d 686 (1964). If CU can be forced to go to trial after this *1113thorough and candid disclosure of its methods, this is the death of consumer ratings: It will be impossible to issue a meaningful consumer review that a band of determined lawyers can’t pick apart in front of a jury. The ultimate losers will be American consumers denied access to independent information about the safety and usefulness of products they buy with their hard-earned dollars. The majority sets a dangerous precedent, and the full court errs grievously by failing to take the case en banc to correct the error.
1. The majority’s analysis is tainted throughout by its failure to articulate, much less apply, a coherent theory of the “independent examination” rale, a key element of the New York Times regime. When a public figure sues his critics for defamation, we “must ‘make an independent examination of the whole record,’ so as to assure ourselves that the judgment does not constitute a forbidden intrusion on the field of free expression.” Id. at 285, 84 S.Ct. 710 (citation omitted). This rale “assigns to judges a constitutional responsibility that cannot be delegated to the trier of fact, whether the factfinding function be perfonned in the particular case by a jury or by a trial judge.” Bose Corp. v. Consumers Union of U.S., Inc., 466 U.S. 485, 501, 104 S.Ct. 1949, 80 L.Ed.2d 502 (1984).1 It is “a rule of federal constitutional law” that “reflects a deeply held conviction that judges ... must exercise such review in order to preserve the precious liberties established and ordained by the Constitution.” Id. at 510-11, 104 S.Ct. 1949.
We have heretofore sensibly assumed that the independent examination rale calls for us to do something more than we woxfid nonnally do. After all, it hardly “preserved] the precious liberties established and ordained by the Constitution” to treat a First Amendment case the same as a slip-and-fall. When we review a jury verdict under New York Times, we don’t “independently examine” the record merely by considering all the evidence — we already do that in every case anyway. See Reeves v. Sanderson Plumbing Prods., Inc., 530 U.S. 133, 150, 120 S.Ct. 2097, 147 L.Ed.2d 105 (2000) (“[I]n entertaining a motion for judgment as a matter of law, the court should review all of the evidence in the record.”). Nor do we merely review de novo the trial judge’s determination whether a jury question exists — again, we already do that in every case. See Janes v. Wal-Mart Stores Inc., 279 F.3d 883, 886 (9th Cir.2002) (“We review a district court’s denial of a motion for judgment as a matter of law de novo .... ”). First Amendment independent examination is an additional protection that the Constitution affords publishers.
Where there are genuine evidentiary disputes over particular factual predicates of actual malice — such as questions about a witness’s credibility' — the issue is for the jury. The judge does not weigh the evidence to find those historical facts, but instead makes a routine sufficiency determination. See Masson v. New Yorker Magazine, Inc., 501 U.S. 496, 520, 111 S.Ct. 2419, 115 L.Ed.2d 447 (1991); Anderson v. Liberty Lobby, Inc., 477 U.S. 242, 255, 106 S.Ct. 2505, 91 L.Ed.2d 202 (1986). After resolving the predicate factual disputes in the plaintiffs favor, however, the judge must take the further step of independently “determining] whether the record establishes actual malice with convincing clarity.” Bose, 466 U.S. at 514, 104 S.Ct. 1949; see Eastwood v. Nat’l Enquirer, Inc., 123 F.3d 1249, 1252 (9th Cir.1997).2 And that process is not a routine sufficiency determination, because it does involve weighing the evidence. If it did *1114not, the independent examination rule would be meaningless.
The majority ignores these principles and instead treats the rule as nothing more than de novo review of the trial judge’s sufficiency ruling. On this theory, independent examination is exactly what we do whenever we consider a trial judge’s determination that a case does not present a triable issue. Of course, de novo review in this sense actually hurts CU because the district court sided with it below.
The majority relies on Bose and Hoffman v. Capital Cities/ABC, Inc., 255 F.3d 1180, 1186 (9th Cir.2001), for the point that independent examination means the same thing as de novo review. But those cases involved a very different use of the term: not de novo review of the lower court’s sufficiency-of-the-evidence determination, but de novo review of the underlying question whether the defendant acted with actual malice. The former type of review is what we do in every case where a party appeals a grant of summary judgment or a denial of judgment as a matter of law. It’s the latter type — weighing the evidence on the ultimate question of actual malice— that constitutes independent examination.3
The majority seems willing to concede that independent examination has some bite after the verdict. But it refuses to “conflate[ ]” this post-verdict standard with the summary judgment one. Am. op. at 1132. By decoupling the two standards, the majority conflicts directly with Supreme Court precedent. As the Court explained in Anderson, “[Tjhere is no issue for trial unless there is sufficient evidence favoring the nonmoving party for a jury to return a verdict for that party.” 477 U.S. at 249, 106 S.Ct. 2505. Because the court must ultimately review a plaintiffs verdict under the independent examination rule, it necessarily follows that it must apply that same standard at summary judgment.
The practical effect of the panel’s decision is that our review for sufficiency at summary judgment is now governed by one standard, while our review after a jury verdict is governed by another. Cases will now often proceed to trial, even though the court can tell ahead of time that the plaintiffs evidence will not support a jury verdict under New York Times. What possible purpose is served by conducting such mock trials? Are they held just in case new evidence materializes midway through trial? Or merely to give the lawyers trial experience? I know of no other context where we allow a plaintiff to get to trial with a factual record that will not support a verdict in his favor.4
*1115The omnipresent danger in defamation suits is that “would-be critics ... may be deterred from voicing their criticism, even though it is believed to be true and even though it is in fact true, because of doubt whether it can be proved in court or fear of the expense of having to do so.” New York Times, 376 U.S. at 279, 84 S.Ct. 710. That a plaintiffs suit ultimately fails after trial is little solace to a defendant crushed by the sheer expense of litigation. For example, in this suit and a similar one filed by Isuzu, see Isuzu Motors Ltd. v. Consumers Union of U.S., Inc., 66 F.Supp.2d 1117 (C.D.Cal.1999), Consumers Union had by September 1999 reportedly spent more than $10 million defending its ratings, while its two adversaries had spent more than $25 million. See John O’Dell, Bruising Tests Await Consumer Reports in Court, L.A. Times, Sept. 19, 1999, at A1. And these are just two of the many lawsuits CU has had to contend with — about a dozen published cases (and who knows how many unpublished ones) involving disgruntled CU reviewees seeking revenge through the courts. Good for lawyers, but not so good for free expression.
It’s no answer to let the chips fall and assume that insurance will cover the tab. Insurance makes costs more predictable, but it doesn’t decrease the amount publishers must pay to do business in the long term. If publishers present a greater liability risk, insurers will charge higher rates. The burden on free speech is no less intolerable.
Even when First Amendment values do not hang in the balance, “[s]ummary judgment procedure is properly regarded not as a disfavored procedural shortcut, but rather as an integral part of the Federal Rules as a whole, which are designed ‘to secure the just, speedy and inexpensive determination of every action.’ ” Celotex Corp. v. Catrett, 477 U.S. 317, 327, 106 S.Ct. 2548, 91 L.Ed.2d 265 (1986) (quoting Fed.R.Civ.P. 1). That principle is all the more vital where a slow and expensive determination will result in self-censorship.
Where a plaintiffs evidence is strong enough to support a verdict, the publisher must stand trial. But there is no sensible reason to impose such onerous costs on a publisher where the plaintiff is doomed to fail. Allowing a plaintiff to get to trial on a record that will not support a verdict under New York Times does not make sense, and if some of the Court’s precedents can be read to suggest this result, the Court would do us all a great favor by explaining that it intended no such absurdity.
2. Had the majority applied the constitutionally required standard, it could not have reached the result it did.5 Suzuki’s case is easily summarized. Its first theory is that CU knew it was probably lying because its employees tried to make the Samurai tip and were happy when they succeeded. The second is that CU purposely avoided the truth by failing to address a potential source of experimental error. Neither of these theories withstands serious scrutiny.
As to the first, the majority relies heavily on the fact that CU switched to the short course after the long course failed to cause any tips. The initial problem with this theory is that CU fully disclosed this in its article. Just so we’re clear what we’re talking about, here’s what CU told the public:
Our regular test program includes a maneuver designed to see how controllable a car remains when a driver is forced to steer sharply — to avoid, say, a child who unexpectedly darted into the *1116road. To simulate that kind of sudden emergency, our drivers run each car through a lane-changing course marked off by traffic cones. The drivers begin their left turn out of the lane 60 feet before the obstacle. They then must steer sharply enough to get back into the lane no more than 60 feet beyond the obstacle.
Under the experienced touch of our drivers, all four utility vehicles got through the course at 52 mph or better. The Suzuki Samurai was actually more maneuverable than the others, since it’s so much smaller and lighter.
With concern about a potential rollover somewhat allayed, a staff member who does not normally drive the course tried to steer the Suzuki around the obstacle. All went well for several runs at moderate speed. Then, on a run at 45 mph, the driver made a slight steering misjudgment: He turned wider than necessary to clear the obstacle, something many ordinary drivers might do in an emergency turn. That forced him to turn back a bit more sharply than our regular testers had. As he turned the steering wheel to the right to get back into lane, the Suzuki teetered to the left. The two right-side tires lifted about a foot off the pavement before the driver was able to bring the vehicle back under control.
Would the other utility vehicles show similar instability if steered the way the Suzuki had been steered?
To find out, we put all the vehicles through a slightly different maneuver. We realigned the cones so that our test drivers had to start the turn 50 feet from the obstacle instead of 60 feet from it. That meant they needed to steer around the obstacle and back into lane in a total of 110 feet rather than the usual 120 feet. We also moved the obstacle three feet farther to the left.
In this more demanding test, the Isuzu Trooper, the Jeep Wrangler, and the Jeep Cherokee began knocking over cones at about 40 mph. But they remained stable. We also tried the Jeep Wrangler without the outriggers. Still stable.
The Suzuki Samurai, by contrast, toppled onto the outriggers when turned through the course at about 40 mph. Without the outriggers, it would have rolled over.
During the period we were testing these vehicles, Suzuki introduced a modified version of the Samurai, a “1988 1/2” model with a softened suspension. We acquired the latest version and ran it through the same accident-avoidance maneuver. It proved even less stable than the Samurai we originally tested. The front right wheel lifted in turns at low speed. And the vehicle rolled onto the outriggers at 38 mph.6
CU thus thoroughly explained both its track design and its testing procedure. It acknowledged that the Samurai did well on the standard course and alerted the reader that CU designed a new course specifically to force a maneuver it had reason to believe the Samurai would fail. Even assuming CU wanted to make the Samurai tip and designed its short course to achieve that result, so what? The fact remains that the Samurai did tip — several times— while every other vehicle run through the same course did not tip even once. This is certainly something consumers would want to know before deciding which of these vehicles to put their families in.
*1117By exposing CU to liability for basing its negative opinion of the Samurai on a fully disclosed testing procedure, the majority runs head-first into a settled First Amendment principle: “ ‘Where a publication sets forth the facts underlying its statement of opinion ... and those facts are true, the Constitution protects that opinion from liability for defamation.’ ” Standing Comm. v. Yagman, 55 F.3d 1430, 1440 (9th Cir.1995) (quoting Lewis v. Time, Inc., 710 F.2d 549, 556 (9th Cir.1983)); see also Partington v. Bugliosi, 56 F.3d 1147, 1156 (9th Cir.1995) (“[W]hen a speaker outlines the factual basis for his conclusion, his statement is protected by the First Amendment.”). The logic behind the rule is straightforward and unassailable: When a publisher prints an opinion but doesn’t state the basis for it, the reader may infer a factual basis that doesn’t exist. See Milkovich v. Lorain Journal Co., 497 U.S. 1, 20-21, 110 S.Ct. 2695, 111 L.Ed.2d 1 (1990). But when a publisher accurately discloses the facts on which he bases his opinion, the reader can gauge for himself whether the factual basis adequately supports the opinion.
This is not a case where a defendant fabricates results and then purports to rely on them — like when NBC secretly strapped rockets to GM trucks with side-mounted fuel tanks to make them explode in collisions. See Jane Hall, Chief of NBC News Quits in Wake of Rigged Crash, L.A. Times, Mar. 3, 1993, at A1. There’s not a shred of evidence that the short course was clandestinely altered to make it more forgiving when the other vehicles went through, or that CU’s drivers intentionally soft-pedaled the runs with the other vehicles to make the Samurai look worse by comparison. The majority’s gripe here is not with CU’s reporting of its tests, but with its experimental design, and that is entirely the wrong focus. If CU had found the Samurai “Not Acceptable” based on tests performed with a Ouija board, most consumers would dismiss its rating as worthless. But, so long as CU disclosed its method, the report would be protected by the First Amendment.
Even aside from the fact that CU fully disclosed its change in course, there is simply nothing improper about what it did. When CU first tested the Samurai on the long course, it didn’t tip — a result consistent with the theory that it is safe, but also consistent with the theory that the long course is not challenging enough to detect the Samurai’s design flaw. CU did what any conscientious researcher would do and redesigned its experiment to focus on the conditions that seemed more likely to detect a flaw. See, e.g., Karl R. Popper, The Logic of Scientific Discovery 45 (Hutchinson rev. ed.1968) (1934) (noting that a scientist, after first detecting a condition, “might try to rearrange his experiments so as to make the effect reproducible”). It then ran the Samurai and several other vehicles through the course and found that the Samurai had a dramatically higher propensity to roll over — it tipped on several occasions while no other vehicle tipped once.
The opinion makes much of the fact that CU had used its long course for testing since 1973, but cites no evidence it was using the course for rollover testing during that time. In fact, the evidence in the record suggests the contrary: None of the 500 vehicles CU had previously tested had rolled over, and it did not even use safety outriggers to protect its drivers from rollovers until the tests in question. Clearly, CU had not previously focused on this particular risk, no doubt because the original test course was designed before SUVs and similar vehicles with high centers of gravity had come into vogue. But after news stories and a NHTSA filing highlighted the Samurai’s rollover propensity, after the vehicle rolled over once during *1118break-in, and after it tipped again following formal testing on the long course, CU had good reason to change focus. Adjusting test procedures to better detect a risk arising from new consumer preferences is hardly reprehensible. It is precisely what one would expect from researchers seriously interested in consumer safety.
Consider an analogy: Assume CU has been testing clothes dryers using a standard load since 1973. It learns that a particular brand of dryer occasionally causes fires. CU dries several standard loads but experiences no adverse results, so it tries a few loads with more flammable materials — say, acetate bed-linens. This is not part of its existing testing protocol, but the materials are ones that consumers have begun to use more often in recent years. If the materials catch fire in the one dryer but not the others, is CU’s test a fraud because it changed its procedure to focus on the new risk?
The relevant question in either case is not whether the new procedure is different from the old, but whether it accurately reflects conditions consumers might occasionally face. And the majority points to no evidence at all on that score. The short course was not some extraordinary challenge that no consumer vehicle could be expected to pass — after all, every other tested vehicle did pass it. CU’s previous use of the long course has no bearing on whether its short course results were fraudulent.
By equating CU’s switch to a short course with knowledge of probable falsity, the majority ignores the realities of experimental design. Scientific inquiry would grind to a halt if researchers couldn’t tinker with their methods as they learned new information about the object of their study. CU’s switch is not even bad science, let alone bad journalism, and certainly not journalism so awful that it loses First Amendment protection. To be probative of actual malice, evidence must have at least some tendency to exclude obvious innocuous explanations for the defendant’s conduct. CU’s switch from a long course to a short course does not.7
That leaves the majority with no direct evidence that CU’s methods were flawed, and only four contemporaneous statements of CU employees — “If you can’t find someone to roll this car, I will,” an unspecified cheer, “That’s it. That looked pretty good,” and “All right Ricky baby” — that, to the majority, show an unfair bias against the Samurai. If CU had a negative view of the Samurai going into the tests, the reasons are not hard to imagine — the Samurai had already been slammed in the press for its rollover propensity, had been the target of a safety complaint filed with NHTSA and had rolled over on its side while a CU employee drove it at only 15 mph. But even if CU’s preconception were entirely arbitrary, these fleeting remarks — three of which are no more than inane schadenfreude — would still be insufficient to support a finding of actual malice.
Actual malice is not bias. “[I]ll will toward the plaintiff, or bad motives, are not elements of the New York Times standard.” Nat'l Ass’n of Letter Carriers v. Austin, 418 U.S. 264, 281, 94 S.Ct. 2770, 41 L.Ed.2d 745 (1974) (internal quotation marks omitted). Speech that is “malicious” only in the sense that the speaker is *1119biased against his target is fully protected; the First Amendment extends beyond “reasoned [and] evenhanded” commentary to the “slashing and one-sided.” Hustler Magazine v. Falwell, 485 U.S. 46, 54, 108 S.Ct. 876, 99 L.Ed.2d 41 (1988).
Bias may support a showing of actual malice; if you bad-mouth someone, the fact that you also don’t like him makes it marginally more likely you’re lying. But bias evidence is not sufficient by itself to support a claim: “[A]ctual malice may not be inferred alone from evidence of personal spite, ill will or intention to injure on the part of the writer.” Harte-Hanks Communications, Inc. v. Connaughton, 491 U.S. 657, 666 n. 7, 109 S.Ct. 2678, 105 L.Ed.2d 562 (1989) (internal quotation marks omitted); see also, e.g., Church of Scientology Int'l v. Behar, 238 F.3d 168, 174 (2d Cir.2001); Shoen v. Shoen, 48 F.3d 412, 417 (9th Cir.1995). If a tabloid sends an investigator to dig up dirt on a celebrity, the fact that it deliberately attempts to find damaging information and is pleased when it succeeds hardly proves it’s lying. Forcing CU to stand trial for statements that show nothing more than bias ignores these settled First Amendment principles.
We don’t hold participants in public debate to Article III standards of impartiality. Nor do we second-guess their investigations as if we were reviewing administrative action under the APA. So long as they don’t he or recklessly disregard the truth, they can be as unfair and one-sided as they want. The majority’s clear and convincing evidence that CU strayed beyond this wide field is shockingly thin. The Second Circuit requires a plaintiff to show that bias was coupled with “an extreme departure from standard investigative techniques.” Behar, 238 F.3d at 174 (emphasis added). For the majority, any departure from past technique will apparently suffice, even one entirely consistent with standard experimental methodology.
What ultimately dooms Suzuki’s case is that, no matter what CU’s motives for choosing the short course, it is undisputed that the Samurai tipped multiple times on that course while the other vehicles tested didn’t tip even once. If the majority has its way, evidence like this — which may make a life-or-death difference to consumers — will be suppressed for fear of precisely this type of lawsuit, waged with gale-force intensity by corporate plaintiffs with bottomless litigation budgets.
3. The majority’s second ground for reversal is weaker still. After CU published its initial report, NHTSA criticized certain aspects of its methodology. CU responded, and the majority finds its response for the most part satisfactory. It concludes, though, that CU fell short by failing to address NHTSA’s critique of its reliance on driver input.
The majority’s analysis suffers from familiar flaws. Even if CU’s reliance on human drivers made its tests less reliable, it disclosed the relevant facts in its report. CU explained in detail its use of human test drivers and the precise maneuvers they performed — right down to the distance between traffic cones. See pp. 1115-1116 supra. Surely these disclosures provide adequate information for readers to make a reasonable assessment of CU’s testing. At the very least, they provide readers familiar with safety testing mechanics — such as NHTSA and Suzuki engineers — the information they need to assess CU’s testing and come up with a public response. CU should not stand trial over a potential source of experimental error that it fully disclosed.
The majority’s holding is flawed for deeper reasons as well. It faults CU for failing to “investigate” and “address” a perceived deficiency in its testing proce*1120dure. Am. op. at 1139. But what was CU supposed to do? Using human drivers obviously introduces a- potential source of experimental error. But it also replicates actual driving conditions and thus captures nuances that might be missed if the experiment were performed by entirely mechanical means. If CU had built some stationary rollover machine instead, Suzuki would have complained that its failure to recreate actual driving conditions made its results unreliable. If CU had built a car-driving robot, Suzuki would have complained that the test didn’t reflect the reactions of a human driver. All experiments have potential sources of error, and many reputable experiments rely on human input. That an experiment can be criticized on these grounds doesn’t make it reckless to rely on the results, even if the criticisms have some validity. Failure to explain away or eliminate all potential sources of error in an experiment cannot be a basis for liability.
What the majority calls “actual malice” is really just one side of a long-running debate over how to test rollover propensity most effectively. CU favors “dynamic” tests using actual drivers, while NHTSA has long preferred “static” tests based on measurements of a vehicle’s geometry. While the majority thinks NHTSA has the better of this scientific debate, Congress was not so sure: In 2000, it passed a statute requiring NHTSA to develop a dynamic rollover testing program. Transportation Recall Enhancement, Accountability, and Documentation (TREAD) Act, Pub.L. No. 106-414, § 12, 114 Stat. 1800, 1806 (2000) (codified at 49 U.S.C. § 30117(c)); see Consumer Information Regulations; Federal Motor Vehicle Safety Standards; Rollover Resistance, 66 Fed.Reg. 35,179, 35,180-81 (request for comments July 3, 2001) (explaining the statute’s history). The TREAD Act followed on the heels of CU’s criticism that NHTSA’s static testing methods, although a “useful predictor of tripped rollover, ... should be used in conjunction with a dynamic stability test using vehicle maneuvers to better predict the risk of untripped rollovers.” 66 Fed.Reg. at 35,180-81. And as NHTSA concedes, the Act “reflects CU’s concern.” Id. at 35,181.
As if that weren’t vindication enough, when NHTSA acted on Congress’s directive, one of the dynamic rollover tests it initially proposed using was the “CU double lane change,” id. at 35,183, the very test the majority now finds so heretical that reliance on it amounts to actual malice. NHTSA identified both the advantages of CU’s test (“face validity”; “probably a good representation of what the public expects of a personal vehicle”; “better represents the dynamics that may result in an untripped rollover”; “maximum speed through the maneuver can be used as part of the vehicle score”; “display[s] the operation of electronic stability control systems”) and its disadvantages (“any vehicle will pass such a test if equipped with tires of sufficiently low traction”; “driving style can strongly influence the test results”; “course layout may cause the steering reversal and roll momentum effect to be more critical for some vehicles than for others”; “a course tuned to one vehicle may not be the worst case for another vehicle to which it is compared”). Although NHTSA ultimately opted for a different set of dynamic rollover tests, those it selected still relied on human drivers and merely involved a lower number of steering inputs (one or two, as opposed to four in CU’s test). See Consumer Information Regulations; Federal Motor Vehicle Safety Standards; Rollover Resistance (Part II), 67 Fed.Reg. 62,528, 62,537-38 (proposed Oct. 7, 2002). NHTSA thus reduced, but did not eliminate, this potential source of error.
*1121Surely these legislative and regulatory developments bear heavily on the disputed issues in the case. Strangely, however, the majority does not even mention them. That Congress specifically repudiated NHTSA’s exclusive preference for static rollover testing shows we cannot blindly accept NHTSA’s criticism of dynamic testing as evidence of actual malice. That NHTSA thought CU’s test was an option worth considering twelve years later shows that it is not so off-the-scientific-map that reliance on it is tantamount to purposeful avoidance of the truth. That NHTSA ultimately chose tests that involve the same source of error, just to a lesser degree, shows that the issue is not nearly so black- and-white as the majority paints it. On a more fundamental level, these developments show that courts have no business wading into this scientific feud over the best way to test rollover propensity. Which method gives the optimal mix of advantages and disadvantages is a point of disagreement among Congress, NHTSA, CU and others. It should be resolved by scientists, policymakers and consumers, not crushing libel verdicts.
The appropriate response to a poorly designed experiment is to conduct a better one and then convince people that your method is more accurate — not to sue the other guy into silence. The point is poignantly made by the facts of this case. For all its kvetching, NHTSA apparently never even performed its own tests on the Samurai — it relied entirely on data from Suzuki. See Am. op. at 1138. Suzuki’s test procedures no doubt had plenty of defects of their own (the most obvious being “company conducting test has financial stake in outcome”). Can we really say the world is better off if the only source of Samurai rollover data is the manufacturer? If consumers are purchasing an SUV they will trust with their lives, aren’t they entitled to know that the Samurai tipped repeatedly in human-driven tests while the other SUVs didn’t tip once?
Groups like CU perform a valuable function in our consumer society, but they suffer from a constant threat of litigation. It’s easy for a jury to second-guess experimental design, and every suit carries the prospect of a massive damages award because the very purpose of a negative review is to convince the reader that the plaintiffs product is not worth buying. It’s hardly surprising, then, that CU attracts lawsuits like moths to a lantern. See, e.g., Bose Corp. v. Consumers Union of U.S., Inc., 466 U.S. 485, 104 S.Ct. 1949, 80 L.Ed.2d 502 (1984) (truly silly dispute over whether sound from plaintiffs speakers wandered “about the room” rather than merely “along the wall” litigated all the way to the Supreme Court). If Suzuki can get to trial on evidence this flimsy, no consumer group in the country will be safe from assault by hordes of handsomely paid lawyers deploying scorched-earth litigation tactics.
The majority’s decision reaches far beyond consumer organizations to virtually any research group that criticizes corporate interests. Many public interest groups are thinly funded, and their experiments are necessarily full of imperfections. Their targets can now stuff unflattering results into a scientific oubliette just by pointing out sources of experimental error and then threatening to sue if the groups don’t “investigate” and “address” them. Today it’s Consumers Union panning the Samurai; tomorrow it could be Greenpeace claiming that some oil refinery is killing fish. The First Amendment doesn’t allow companies to squelch their scientific detractors on evidence as thin as Suzuki’s. The majority’s legal regime cannot coexist with our tradition of robust scientific debate on matters of public concern.
4. Judge Graber focuses on two particular statements, but succeeds only in high*1122lighting how deficient Suzuki’s case is. She first faults CU for reporting that it developed the short course “because [it] discovered” the Samurai’s rollover propensity, when in fact it developed it “in order to create” that propensity. Concurrence at 1141 (emphasis omitted).
I don’t understand how CU could have “createfd]” the Samurai’s rollover propensity. A propensity is something innate to the object. Is Judge Graber suggesting that CU somehow modified the vehicle to make it more likely to tip? There is no evidence of that whatsoever. What CU did was put the Samurai through a test that demonstrated its innate tendency to tip, a propensity it had as a result of the way it was built. CU designed a course to develop evidence of the vehicle’s propensity to roll over. It could not and did not create the propensity itself.
That problem aside, there is simply nothing false about CU’s statement. There’s no dispute why CU built the short course: As it thoroughly explained in its article, it was trying to replicate the conditions that caused the Samurai to tip when a CU employee turned too sharply on the long course. Suzuki doesn’t dispute this. CU switched to the short course “because [it] discovered” a propensity to roll over during that particular maneuver, a maneuver none of the drivers had previously performed because they had all driven the long course as designed. CU’s statement is an entirely accurate description of its reason for developing the short course.8
Judge Graber also complains that CU said the Samurai rolls over “easily” when, in fact, it had to be “coaxed.” Concurrence at 1141. She apparently fears that readers might assume the Samurai rolled over with no “coaxing” at all. But no one could be so misled. Even tremendously unsafe vehicles roll over only in extreme maneuvers. No one reading that the Samurai rolls over “easily” would infer it routinely flops over with no human intervention.
Saying a consumer vehicle rolls over “easily” is like saying a particular NBA player is “terrible.” The adjective is inherently relative because the reference group is already extreme. A “terrible” professional athlete is one who’s terrible relative to his peers, not relative to the population at large or a class of third-graders. Likewise, the mental image conjured up when one hears the Samurai rolls over “easily” is not of some oopsy-daisy clown car, but a vehicle that rolls over easily relative to other vehicles in its class. And, since all vehicles require some coaxing to roll over, Judge Graber’s fear' — that readers might be duped into thinking they’ll come out of the supermarket to find that their Samurai had flopped itself over in the middle of the parking lot — is a specter of her own creation. Even ignoring the rollover during break-in and the tip on the long course — not to mention the NHTSA complaint and the media coverage — CU was plainly justified in concluding that, given the alternatives, the Samurai’s rollover propensity was more than what safety-conscious consumers should be willing to bear. And that’s exactly what any reader would understand it to have said.
*1123Moreover, the word “easily” cannot be plucked out of context. It must be read in light of CU’s article (to which the anniversary issue obviously referred). CU explained exactly how “easily” the Samurai tipped over, and what CU did to reach its conclusion. It did not secretly put lead weights in the roof to raise the center of gravity; it did not use narrow tires to cause instability; it did not jimmy the suspension. It did exactly what it said it did — -it put the Samurai through the same course as the other vehicles, and the Samurai tipped while the others did not. CU did not ask consumers to accept its claim at face value; it gave them the data on which it based its conclusion so they could decide for themselves whether or not the Samurai tipped over too easily.
Even if CU’s reporting were somehow inaccurate or misleading, there are any number of legal doctrines that protect it. (“Rhetorical hyperbole” comes to mind, see Letter Carriers, 418 U.S. at 285-86, 94 S.Ct. 2770.) But it’s not. CU had adequate grounds to conclude that the Samurai rolled over easily compared to similar vehicles, and that’s all it said.
Judge Graber doubtless perused the record with great care in search of something to support Suzuki’s case. That she could come up with nothing better than these two statements speaks volumes about the overall accuracy and candor of CU’s reporting.
I have read CU’s review of the Samurai and Suzuki’s criticism of its methodology. After all that, I can only say I would long hesitate before letting anyone I care about drive or ride in one of these vehicles. If Suzuki wanted me to disregard CU’s conclusions, it should have taken the money it spent on this lawsuit and hired an independent agency to run tests showing that CU’s criticisms are unfounded. It could also have tried to improve its product to moot criticism in the future. But, until today, I had thought the one option not available to a company in Suzuki’s position was to use its vast financial resources to drag its critics through the gauntlet of our immensely expensive litigation machine. I continue to hope I’m right, or this will be a sad day indeed for consumer organizations and those who rely on them for information vital to their health and safety.
*1124APPENDIX
[[Image here]]
*1125[[Image here]]
*1126[[Image here]]
[[Image here]]
*1127OPINION
TASHIMA, Circuit Judge.In 1988, Appellee Consumers Union of United States, Inc. (CU), published a story in its magazine Consumer Reports, in which it rated the Suzuki Samurai “Not Acceptable” based on its propensity to roll over during accident avoidance tests. Since that time, CU has pubhcly referred to the negative Samurai rating in various fora, most prominently in the 60th Anniversary issue of Consumer Reports, published in 1996. Appellant Suzuki Motor Corporation (Suzuki), the manufacturer of the Samurai, has chahenged the validity of CU’s Samurai test and, on the heels of the 60th Anniversary issue, brought this action against CU alleging product disparagement. CU’s motion for summary judgment was granted by the district court, which held that a reasonable jury could not conclude by clear and convincing evidence that CU had acted with actual malice. We have jurisdiction under 28 U.S.C. § 1291, and we reverse and remand for further proceedings.
I. BACKGROUND
CU is a nonprofit corporation that engages in comparative testing and evaluation of consumer products and services, the results of which are published in the magazine Consumer Reports. In order to provide buying and safety advice to automobile purchasers, CU’s Automotive Testing Division (ATD) tests approximately 40 cars and other vehicles each year.
CU tested the Samurai in 1988.1 The Samurai, a sport utility vehicle (SUV) manufactured by Suzuki, was introduced in the United States in 1985. By 1988, approximately 150,000 Samurais had been sold. Although it had received some favorable reviews from the automotive press, the Samurai had also been the subject of news stories that highlighted its instability and propensity to tip over. In February 1988, the Center for Auto Safety filed a petition with the National Highway Traffic Safety Administration (NHTSA) to open an investigation into an alleged safety defect of the Samurai. The petition was denied, although the NHTSA emphasized that the denial was not an endorsement of the safety performance of a vehicle.
A. April 20, 1988: Long Course Testing
On April 20, 1988, the ATD tested the Samurai, along with the Jeep Cherokee, Isuzu Trooper II, and Jeep Wrangler, on its standard long course, a double lane-change avoidance maneuver test course that CU had used since 1973.2 The long course was designed to replicate an emergency situation in which a driver suddenly steers a vehicle left into the opposing lane, to avoid an obstacle, and quickly back into the original lane to avoid oncoming traffic. Several CU personnel were in attendance during the April 20 testing, including Rob*1128ert Knoll, the head of the ATD, Dr. R. David Pittle, CU’s Technical Director and Senior Vice-President, and Irwin Landau, the Editorial Director of Consumer Reports, who had been assigned as the initial writer and editor of the Samurai article. Pittle had invited Landau to attend the testing because he thought that they might witness a tip-up of the Samurai.
CU driver Kevin Sheehan drove the Samurai first, putting it through the long course 16 times at speeds reaching over 50 miles per hour. The Samurai that Shee-han drove was equipped with outriggers, which essentially act as training wheels to prevent the car from tipping over completely.3 During Sheehan’s runs, the Samurai did not tip over, prompting Sheehan to make the following evaluation of the car: “rubbery, slow response, rocks a bit, but never felt like it would tip over.” In the Avoidance Maneuver Data Summary, Sheehan rated the Samurai as highly as or better than other vehicles tested that day.
After Sheehan had completed his testing, CU removed the outriggers. CU driver Rick Small then drove the Samurai through the long course 21 times at speeds similar to those achieved by Sheehan. Again, there were no tip-ups. In his driver log, Small stated: “steering is slow, but it works — responds well and corrects quickly, leans normally, snaps back. Confidence fairly high. No real problem.” On the basis of his test drives, Small rated the Samurai higher than the other three vehicles tested that day.
According to testimony by former CU employee Ron Denison, at some point during the long-course testing, which had not demonstrated any tip-ups of the Samurai, Landau told Sheehan: “If you can’t find someone to roll this car, I will.”
After Sheehan and Small had completed their test runs, Pittle, who was not a test driver, began to drive the Samurai through the long course. According to Pittle, he did so because he had never driven a small SUV and wanted to get a feel for how it handled through the course. Pittle took the car though the course 10 times, achieving a top speed of 49 miles per hour. On Pittle’s tenth run, the Samurai tipped up on two wheels. Pittle stated that he did not purposefully cause the Samurai to tip up and that it was a startling and unexpected occurrence. When Pittle tipped the car, one onlooker yelled, “yeah!,” while another shouted, “I think I got that, I think I got that.”
B. April 26,1988: Short-Course Testing
After the long-course testing, Knoll redesigned CU’s avoidance course to replicate the situation that caused the Pittle tip-up. This new modified short course had a reduced distance for the first lane change, and the obstacle to be avoided was moved three feet to the left.
Sheehan, who was afraid to drive the Samurai through the short course, was replaced by CU driver Fred Wood. Wood, who drove the vehicle with outriggers, made 15 runs through the course. On the fifteenth run, the Samurai tipped up onto the outriggers. After this last run, Knoll is heard on the test videotape saying: “That’s it. That looked pretty good.” Knoll later acknowledged that he was “relieved” that the Samurai tipped up during short-course testing.
Small then drove the car through the short course. On his second run, the Samurai tipped up onto the outriggers at a speed of 40 miles per hour. After the tip-*1129up occurred, CU technician Joseph Nappi can be heard on the test videotape saying, “[a]ll right Ricky baby.”4 That same day, CU also put the Jeep Wrangler and Isuzu Trooper II through the short course with no tip-ups reported.
C. May 12, 1988: Additional Short-Course Testing
Further testing was scheduled on the short course for May 12, 1988. The parties offer conflicting justifications for why this set of tests was necessary. Suzuki suggests that the additional testing was scheduled for the purpose of shooting video footage of the Samurai for a subsequent press conference, at which it had already been decided that CU would rate the Samurai “Not Acceptable.” CU contends that the testing was held to evaluate the performance of the newly released 1988 Samurai, which included suspension modifications that distinguished it from the 1988 version tested previously.
Small was the first to drive the Samurai through the short course on this day, tipping over onto the outriggers on his fifth run. Following Small, Wood tipped the Samurai on his second run. Pittle was watching the test runs and, prior to witnessing a tip up, stated: “Can’t you just see it, we get no lift off the ground. Oh God.” CU also put a Jeep Wrangler, Jeep Cherokee, and Ford Festiva sedan through the short course on May 12. None of these vehicles tipped up.
D. June 2,1988: Press Conference
CU held a press conference on June 2, 1988, at which it announced that the Samurai had shown a propensity to roll over in CU’s tests and that it would be rated “Not Acceptable” in an article to appear in the July 1988 issue of Consumer Reports. During the press conference, Pittle stated that, based on CU’s testing, the Samurai had an “unusually high propensity to roll over while performing an accident avoidance maneuver that could be demanded suddenly of any driver during routine driving.” Pittle further described the short course as “benign,” involving only “very limited steering inputs,” a characterization he later conceded was “not accurate.” Pit-tie also stated that the other tested vehicles had made it through the short course with a “yawn,” a statement that Suzuki contends is at odds with the fact that the Isuzu Trooper II failed the course in three out of four runs by hitting cones.5
E. July 1988: Article Publication
CU’s negative rating of the Samurai was detailed in an article entitled, “Warning: The Suzuki rolls over too easily,” published in the July 1988 issue of Consumer Reports. The article was initially written by Landau, although drafts were reviewed by Pittle, Knoll, the ATD staff, CU’s President Rhoda Karpatkin, CU’s Technical Department and Library, and legal counsel.
*1130The article described the steps CU took to test the Samurai, Jeep Wrangler, Isuzu Trooper II, and Jeep Cherokee. It began by recounting the incident in which an ATD staff member rolled the Samurai over on its side during the evaluation and break-in process. After detailing other evidence of the Samurai’s safety problems, the article then described the long-course testing, noting that Pittle, “a staff member who does not normally drive the course,” tipped the Samurai at 45 miles per hour after making “a slight steering misjudgment” that should not have “put daylight under the tires of any car.” The article continued by highlighting the results of both short-course tests, concluding that “that Suzuki Samurai is so likely to roll over during a maneuver that could be demanded of any car at any time that it is unfit for its intended use. We therefore judge it Not Acceptable.”
On June 2, 1988, CU submitted to NHTSA a copy of the article, the videotape from the press conference, and a diagram of its short course in support of a petition to establish a minimum stability standard to protect against unreasonable risk of rollover.
F. NHTSA Report
On September 8, 1988, NHTSA issued a decision denying the motor vehicle defect petition filed earlier by the Center for Auto Safety. In its analysis, the NHTSA stated that “the rollover crash involvement of the Samurai appears to be within the range of most other light utility vehicles.”6 Denial of Motor Vehicle Defect Petitions, 53 Fed.Reg. 34,866 (Sept. 8, 1988). It concluded that the Ford Bronco II “was found to have more than 3 times the first event rollover ... involvement as the Samurai” and that “the Samurai had a first event rollover involvement corresponding to the [Chevrolet] S-10 Blazer.” Id.
NHTSA’s opinion also criticized CU’s testing protocols, stating as follows: “The existing test procedures for assessing the rollover propensity of vehicles are unsatisfactory because they do not provide for repeatable, reproducible results, and there are no accepted performance criteria. The testing appears to rely on the skill and influence of the driver and the presumption that the vehicle suspension, tire, and road surface characteristics will remain constant throughout the testing.” Id. at 34,867. NHTSA concluded by stating that, although the CU testing results were “cause for some concern,” “the test procedures do not have a scientific basis and cannot be linked to real-world crash avoidance needs, or actual crash data. Using the same procedures, probably any light utility vehicle could be made to roll over under the right conditions and driver input.” 7 Id. CU issued a subsequent article in Consumer Reports criticizing the NHTSA decision.
G. 1988-1996: CU’s Republications of the Samurai Report and Further Rollover Claims
Between 1988 and 1996, CU republished references to the 1988 Samurai rating on at least 24 separate occasions in Consumer Reports, CU’s annual buying guide, and other editions of CU’s car books. During this time, CU states, several events bolstered its belief in the correctness of its *1131“Not Acceptable” rating: a 1988 England-based Consumers’ Association article that buttressed the Samurai rollover claim; a 1988 lawsuit filed by seven state Attorneys General charging Suzuki with false and misleading advertising regarding the Samurai’s rollover potential (the case settled); the decision in Malautea v. Suzuki Motor Corp., 148 F.R.D. 362, 375 (S.D.Ga.1991), in which the court suggested that Suzuki knew of the Samurai’s rollover propensity and did nothing to correct it; two separate multimillion dollar verdicts in the case of Rodriguez v. Suzuki Motor Co., where juries determined that the Samurai was unreasonably dangerous due to its rollover propensity (both verdicts were reversed and the case settled); the disclosure of documents from the Malautea and Rodriguez cases suggesting that Suzuki knew of the Samurai’s rollover propensity; and eight years of further SUV testing by CU during which time only the Samurai in 1988 and Isuzu Trooper in 1996 tipped up.
H. January 1996: 60th Anniversary Issue of Consumer Reports
In its 60th Anniversary issue of Consumer Reports, published in January 1996, CU set forth a chronology that contained a picture of the Samurai tilted on two wheels, with the following caption:
1986 CU buys its own auto test track in rural Connecticut. Two years later, based on tests conducted there, CONSUMER REPORTS discovers the Suzuki Samurai easily rolls over in turns and rates it Not Acceptable. Sales of the Samurai dwindle away. Since 1936, dozens of products, from chemistry sets and toasters to power mowers and child safety seats, have been identified as safety hazards and rated Not Acceptable.
The same January 1996 issue also contained a section entitled “Memo to Members,” in which CU’s President stated that “we still find products that are unsafe: From kerosene heaters to the Suzuki Samurai to child safety seats, CONSUMER REPORTS has called them out — and our work goes on.”8
I. April 1996: Suzuki Files This Action
On April 11, 1996, Suzuki filed the instant action alleging that CU’s ongoing publication of the negative Samurai rating constituted product disparagement.9 After discovery, CU moved for summary judgment, challenging the sufficiency of Suzuki’s evidence that CU had acted with actual malice in its reporting on the Samurai. The district court granted CU’s motion and entered judgment in its favor. Suzuki timely appealed.
II. STANDARD OF REVIEW
A grant of summary judgment is reviewed de novo. Devereaux v. Abbey, 263 F.3d 1070, 1074 (9th Cir.2001) (en banc). This court’s review is governed by the same standard used by the trial court under Federal Rule of Civil Procedure 56(c). Adcock v. Chrysler Corp., 166 F.3d 1290, 1292 (9th Cir.1999). We must therefore determine, viewing the evidence in the light most favorable to the nonmoving party, whether there are any genuine issues of material fact and whether the district court correctly applied the relevant sub*1132stantive law. Devereaux, 263 F.3d at 1074.
Because this case implicates the First Amendment protections of a media defendant in the context of product disparagement, “[t]he appropriate summary judgment question is whether a reasonable jury could find, by clear and convincing evidence, that [the plaintiff] has shown actual malice.” Kaelin v. Globe Communications Corp., 162 F.3d 1036, 1039 (9th Cir.1998). In answering this question, we “must draw all justifiable inferences in favor of the nonmoving party, including questions of credibility and of the weight to be accorded particular evidence.” Masson v. New Yorker Magazine, Inc., 501 U.S. 496, 520, 111 S.Ct. 2419, 115 L.Ed.2d 447 (1991). “[T]he plaintiff, to survive the defendant’s motion, need only present evidence from which a jury might return a verdict in [its] favor. If[the plaintiff] does so, there is a genuine issue of fact that requires a trial.” Anderson v. Liberty Lobby, Inc., 477 U.S. 242, 257, 106 S.Ct. 2505, 91 L.Ed.2d 202 (1986).
The dissent contends that, by applying the well-established summary judgment rules to the actual malice issue on summary judgment, we offend the “independent examination” standard of New York Times Co. v. Sullivan, 376 U.S. 254, 84 S.Ct. 710, 11 L.Ed.2d 686 (1964). This argument, however, conflates the summary judgment standard of review with application of the New York Times standard and, as a result, impermissibly weighs the evidence at the summary judgment stage. Under the independent examination rule, we “exercise our independent judgment” in evaluating the lower court’s opinion, rather than granting it any deference. See Bose Corp. v. Consumers Union of United States, Inc., 466 U.S. 485, 514, 104 S.Ct. 1949, 80 L.Ed.2d 502 (1984). In other words, we review the district court’s decision de novo. See Hoffman v. Capital Cities/ABC, Inc., 255 F.3d 1180, 1186 (9th Cir.2001) (“We review the district court’s finding of actual malice de novo.”); Eastwood v. Nat’l Enquirer, Inc., 123 F.3d 1249, 1252 (9th Cir.1997) (“First Amendment questions of ‘constitutional fact’ compel us to conduct a de novo review.” (internal quotation marks and brackets omitted) (quoting Bose, 466 U.S. at 508 n. 27, 104 S.Ct. 1949)). This does not mean, however, contrary to the dissent’s suggestion, that in the process of exercising our independent judgment, we jettison the procedural rules governing summary judgment when reviewing the grant of summary judgment in First Amendment cases. While it is true that we must independently examine the record when reviewing the grant of summary judgment, starting with Anderson v. Liberty Lobby, the Court has been clear that we apply the well-established summary judgment rules to review of the actual malice issue:
Consequently, where the New York Times “clear and convincing” evidence requirement applies, the trial judge’s summary judgment inquiry as to whether a genuine issue exists will be whether the evidence presented is such that a jury applying that evidentiary standard could reasonably find for either the plaintiff or the defendant. Thus, where the factual dispute concerns actual malice, clearly a material issue in a New York Times case, the appropriate summary judgment question will be whether the evidence in the record could support a reasonable jury finding either that the plaintiff has shown actual malice by clear and convincing evidence or that the plaintiff has not.
Anderson, 477 U.S. at 255-56, 106 S.Ct. 2505. See also Masson, 501 U.S. at 520, 111 S.Ct. 2419 (“On summary judgment, we must draw all justifiable inferences in favor of the nonmoving party, including questions of credibility and of the weight *1133to be accorded particular evidence.”). And, as we have noted above, our case law also recognizes the application of the normal summary judgment standards to the actual malice issue. See Solano v. Playgirl, Inc., 292 F.3d 1078, 1082, 1087 (9th Cir.), cert. denied, — U.S. —, 123 S.Ct. 557, 154 L.Ed.2d 443 (2002); Kaelin, 162 F.3d at 1039. Thus, contrary to the dissent’s suggestion, the independent examination rule of New York Times is consistent with our well-established procedural rules governing review of summary judgment motions on the actual malice issue and does not require us to discard the procedural rules designed to preclude the resolution of disputed factual issues at the summary judgment stage.10
Thus, although the dissent provides a plausible view of the evidence — that CU acted in good faith as a skeptical consumer watchdog should — it is not our role, at this stage, to take sides in this way. As we discuss below, there is also another plausible view of the summary judgment record — that CU “rigged” a test to achieve a predetermined result in order to serve its own pecuniary interests. Because a jury would be entitled to believe the latter view of the evidence, Suzuki’s case survives summary judgment, even applying the independent examination standard.
III. ANALYSIS
The parties do not dispute that for Suzuki to recover in this case, it must, as a public-figure plaintiff, prove by clear and convincing evidence that CU published disparaging statements about the Samurai with actual malice.11 See Unelko Corp. v. Rooney, 912 F.2d 1049, 1057-58 (9th Cir.1990) (stating that claims for product disparagement “are subject to the same first amendment requirements that govern actions for defamation”); see also Isuzu Motors Ltd. v. Consumers Union of United States, Inc., 66 F.Supp.2d 1117, 1124 (C.D.Cal.1999); Melaleuca, Inc. v. Clark, 66 Cal.App.4th 1344, 78 Cal.Rptr.2d 627, 637 (1998); cf. Bose, 466 U.S. at 513, 104 *1134S.Ct. 1949 (assuming without deciding that proof of actual malice is required in a product disparagement action brought by a public-figure plaintiff against a media defendant). Actual malice requires a showing that the defendant made a false statement “with knowledge that the statement was false or with reckless disregard as to whether or not it was true.” Hustler Magazine v. Falwell, 485 U.S. 46, 56, 108 S.Ct. 876, 99 L.Ed.2d 41 (1988); accord Harte-Hanks Communications, Inc. v. Connaughton, 491 U.S. 657, 667, 109 S.Ct. 2678, 105 L.Ed.2d 562 (1989).
This Court has interpreted Harte-Hanks as providing two different tests for proving reckless disregard. Masson v. New Yorker Magazine, Inc., 960 F.2d 896, 900 (9th Cir.1992) (Masson II):
Where the jury has proof that a publisher “actually had a high degree of awareness of probable falsity,” that alone will establish that it “in fact entertained serious doubts as to the truth of [its] publication.” Where such direct proof is missing, the jury may nevertheless infer that the publisher was aware of the falsity if it finds that there were “obvious reasons to doubt” the accuracy of the story, and that the defendant did not act reasonably in dispelling those doubts .... As Harte-Hanks points out, “[although failure to investigate will not alone support a finding of actual malice, the purposeful avoidance of the truth is in a different category.”
Id. (quoting Harte-Hanks, 491 U.S. at 688, 692, 109 S.Ct. 2678) (brackets in the original).
Suzuki argues that the district court erred in concluding that a reasonable jury could not find that CU’s statements regarding the Samurai met the actual malice standard. It contends that the evidence is sufficient to sustain a jury verdict under either test articulated in Masson II for proving CU’s reckless disregard of the truth.
A. Awareness of Probable Falsity
Suzuki first contends that a reasonable jury could find by clear and convincing evidence that CU had a high degree of awareness of the probable falsity of its statements about the Samurai. It argues that there is a sufficient evidentiary basis showing that CU knew that the Samurai did not tip up more easily than other SUVs and essentially rigged its testing to produce a predetermined result.
At the outset, CU raises a general challenge to Suzuki’s argument, suggesting that the overwhelming weight of the evidence demonstrates the skill and dedication of CU in researching and publishing the Samurai story, and therefore militates against any finding of actual malice. CU assails Suzuki for taking “a few facts out of context, which it pieces together in a contrived and inherently implausible fashion without any supporting evidence.” It argues that this court should reject Suzuki’s arguments as conjectural and places great weight on its own assertions that it absolutely believed in the truth of its statements.
In the summary judgment context, this argument is unconvincing. It is true that CU has offered evidence of its accuracy in reporting the Samurai story — or, at least, in its subjective belief that it accurately reported the story. But its characterization of its own evidence as overwhelming and its disparagement of Suzuki’s evidence as out of context begs the question that we must resolve. The fact that CU employees believed in the truth of their negative statements about the Samurai cannot, by itself, defeat summary judgment. See St. Amant v. Thompson, 390 U.S. 727, 732, 88 S.Ct. 1323, 20 L.Ed.2d 262 (1968) (“The defendant in a defamation action ... can*1135not ... automatically insure a favorable verdict by testifying that he published with a belief that the statements were true.”); Solano, 292 F.3d at 1087 (“we have yet to see a defendant who admits entertaining serious doubt about the authenticity of an article it published”). Rather, the issue is whether there is adequate evidence to support the contrary view — namely, that behind the veneer of accuracy, CU was disseminating the Samurai story with knowledge, or reckless disregard, of its falsity. While it may be true that CU’s evidence of meticulous reporting ultimately has more weight than Suzuki’s evidence of actual malice, that is not a question to be resolved here. See Anderson, 477 U.S. at 249, 106 S.Ct. 2505.
Turning to the evidentiary basis for Suzuki’s claims, Suzuki contends that there is sufficient evidence that CU “rigged” the tests to produce the rollover result, demonstrating its awareness of the probable falsity of the negative Samurai rating. In support of this argument, Suzuki highlights the testimony of Denison, who stated that Landau, after witnessing Sheehan and Small’s clean runs through the long course, said: “If you can’t find someone to roll this car, I will.” After this comment was allegedly made (and after Sheehan and Small had taken the Samurai through 37 long-course runs without incident), Suzuki notes that Pittle, who was not a CU test driver, decided to drive the Samurai, tipping the vehicle after 10 runs to a cheer from a CU onlooker.12 Suzuki also points to the fact that CU modified the long course, which it had used since 1973, to replicate the Pittle situation and then resumed its testing of the Samurai until it tipped up — again eliciting statements that can be construed as expressions of satisfaction (“That’s it. That looked pretty good.” “All right Ricky baby.”).
This evidence is adequate to preclude summary judgment. A reasonable jury could find by clear and convincing evidence that CU sought to produce a predetermined result in the Samurai test. The timing of the course modification, the fact that the Suzuki was tested repeatedly until it tipped, and the reactions of CU employees all support such an inference. A permissible inference of reckless disregard follows from this evidence of “rigged” testing — if CU modified the course in order to cause a rollover, a reasonable jury could find that the truthfulness of any subsequent reports was vitiated.
Although it suggests that the facts have been taken out of context, CU does not seriously challenge Suzuki’s argument of rigging. CU does question Suzuki’s reb-anee on the Denison testimony, arguing that he had a high regard for CU’s ethical standards and did not believe that anyone involved in CU’s testing of the Samurai rigged the results. Denison did testify, however, that while he believed CU was honest 99.9 percent of the time, the 0.1 percent that he was excluding was the 1988 test of the Samurai. Moreover, although CU has its own interpretation of why it modified the long course and retested the Samurai, we, of course, cannot credit that interpretation over Suzuki’s at the summary judgment stage. The district court did not give adequate credit to this evidence of test-rigging.
Suzuki also contends that evidence of CU’s avaricious motives supports an inference of actual malice. It suggests that the evidence reveals that CU was financially overextended due to capital investments in the period leading up to the Samurai report and needed a blockbuster story to raise CU’s profile and increase fundraising revenues. It is not disputed that, at the *1136time of the Samurai story, CU had incurred substantial debt for a new headquarters and that CU has used the Samurai story in its fundraising solicitations.
CU characterizes Suzuki’s financial motive argument as an “unsupported accusation” and states that, instead of increasing revenues, CU’s pre-publication press conference about the Samurai test actually decreased its sales of Consumer Reports. CU further contends that evidence of financial motive does not support a finding of actual malice under the relevant case law. See Harte-Hanks, 491 U.S. at 667, 109 S.Ct. 2678 (“Nor can the fact that the defendant published the defamatory material in order to increase its profits suffice to prove actual malice.”). The district court adopted this argument in granting summary judgment.
We agree with Suzuki, however, that the district court erred in this regard. There is sufficient circumstantial evidence of a financial motive to support the ultimate conclusion of actual malice. While CU is correct that financial motive cannot, by itself, prove actual malice, it nonetheless is a relevant factor bearing on the actual malice inquiry. See Harte-Hanks, 491 U.S. at 668, 109 S.Ct. 2678; Kaelin, 162 F.3d at 1042. The evidence of financial motive dovetails with the evidence of test-rigging described above. The fact that CU needed to boost its revenues to complete its capital campaign lends credence to Suzuki’s contention that CU rigged the Samurai testing to produce the predetermined rollover result.
We conclude that the evidence of motive and test-rigging, in combination, is sufficient to preclude summary judgment and therefore requires reversal.13
B. Purposeful Avoidance
Suzuki also argues that there was sufficient evidence for a jury to have concluded under Masson II that, in the face of obvious reasons to doubt the accuracy of its Samurai story, CU did not act reasonably in dispelling those doubts, thereby raising the inference that CU knew of the story’s falsity. Under this approach, “[i]t is not ... the failure to act reasonably in itself that establishes malice; that failure is only a link in the chain of inferences that could (but need not) lead a jury to conclude that the publisher failed to conduct an investigation because it was already pretty much aware of the falsity.” Masson II, 960 F.2d at 900. The central inquiry is whether the evidence discloses that a defendant purposefully avoided the truth. Id.
First, Suzuki claims that CU had reason to doubt its assertion that the Samurai’s rollover propensity warranted a “Not Acceptable” rating. In particular, Suzuki points to the NHTSA decision issued in September 1988 indicating that the Ford Bronco II had a three times greater rollover record than the Samurai, which had a rollover record corresponding to the Chevrolet S-10 Blazer. Additionally, CU learned in 1989 that the Samurai’s rollover rate was less than the rollover rate of the *1137Nissan Pathfinder, Toyota 4 Runner, Jeep CJ-7, and Ford Bronco II. Further, in 1992, the Insurance Institute for Highway Safety produced a report showing that the Samurai had a rollover rate lower than that of many other SUVs.
Second, Suzuki contends that CU had sufficient reasons to doubt the validity of the testing procedures it used to determine the negative Samurai rating. Here Suzuki again relies on the NHTSA study criticizing CU’s testing procedures for failing to provide a basis for repeatable results and being overly dependent on driver influence. Suzuki also emphasizes the British Department of Transport study echoing these conclusions, as well as statements by Knoll and Sheehan suggesting that at least some CU personnel acknowledged that its testing procedures were overly driver-influenced. Suzuki asserts that, despite these indications that the Samurai rating rested on questionable data and suspect methodological premises, CU took no steps to engage in further investigation, thereby reinforcing the inference of purposeful avoidance. Suzuki faults CU for failing to incorporate instruments into its testing that would record how the driver was steering the vehicle and for never evaluating its test results against real-world crash data. It suggests that CU’s failure to do so violated accepted journalistic standards14 and raises the inference that CU ignored contrary evidence that would confirm the falsity of its claims about the Samurai. The district court rejected much of this evidence, stating particularly that the NHTSA study was not entitled to greater weight than any other study or opinion regarding testing methods and therefore could not support a claim of actual malice.
In response to the NHTSA report, CU published an article in the November 1988 issue of Consumer Reports that addressed the NHTSA’s critique of CU’s negative Samurai rating. With respect to the issue of the Samurai’s rollover rate, CU stated:
According to NHTSA’s own Crash Avoidance Research Data file, however, the Suzuki Samurai’s rate of rollover in single-vehicle accidents is more than double the average for all sport-utility vehicles. In 1986, the most recent year for which there are figures, the Suzuki rolled over in 64 percent of all single-vehicle Suzuki accidents reported in this data base. The only vehicle that came close to the Samurai in rollover involvement is the now-discontinued Jeep CJ-5 (49 percent). By contrast, the rollover rate for full-sized sedans was only 8 percent.
NHTSA appears to have relied not on its statistics on rollover rates for single-vehicle accidents but on a different data base, one that includes only rollovers involving a fatality. Elsewhere, ... [the NHTSA] notes that the Samurai was involved in six fatal rollovers per 100,000 vehicles on the road, a record the agency compares favorably with that of the Ford Bronco II — 19 fatal rollovers per 100,000 vehicles on the road.
CU has learned that the overall rollover rate for the Bronco II is high— about the same as that for the Jeep CJ-5 — but not nearly as high as the Su*1138zuki’s. The higher number of fatalities in Bronco II rollovers could come about for a number of reasons. The Suzuki rolled over at a relatively low speed in our accident-avoidance tests; if Bronco II rollovers occurred at higher speeds, one would expect more fatalities per rollover. One would also expect more fatalities if the Bronco II were driven more miles, on average, than the Suzuki.
Regarding the NHTSA’s criticisms of CU’s testing protocols, the article went on to state:
NHTSA did no independent testing of the Suzuki’s rollover propensity. Rather, it uncritically accepted Suzuki’s data, saying Suzuki “demonstrated that the Samurai satisfactorily completed industry accepted ... tests which might be used to assess a vehicle’s rollover propensity.”
But there are no industry-accepted tests for rollover propensity — a point NHTSA itself makes elsewhere in its letter....
[One] test performed for Suzuki looks superficially like an avoidance maneuver, since the car was run through a slalom course. But in a realistic avoidance maneuver, a car is steered first to the left and then back to the right immediately. In the Suzuki test, the car was steered to the left, then straightened and allowed to recover before returning to the right lane. That is a simple lane-changing maneuver, not an accident-avoidance maneuver.
NHTSA also adopted as its own another of Suzuki’s arguments: Using the accident-avoidance maneuver developed by CU, the agency stated, “probably any light utility vehicle could be made to roll over.” But NHTSA offered no evidence or independent test results to support such speculation. In fact, no vehicle other than the Suzuki has rolled over in the 10 years we’ve tested for accident avoidance.
The critical inquiry under Masson II is whether CU failed to act reasonably in investigating and responding to contrary studies in a manner that suggested it was attempting purposefully to avoid discovering the truth of the matter. In general, the analysis conducted and published by CU in response to the NHTSA study is not indicative of purposeful avoidance. To the contrary, in the November 1988 article, CU challenged the NHTSA report head on, stating its disagreement in detail and supporting its alternative conclusions with substantive justifications. To the extent that there were contrary rollover statistics, CU analyzed them and explained why they did not warrant a conclusion at odds with its initial assessment of the Samurai. In response to the NHTSA’s critique of CU’s testing, CU argued that the Suzuki tests upon which the NHTSA relied were flawed. While Suzuki may disagree with CU’s discussion of the rollover statistics or its criticisms of Suzuki’s own accident avoidance tests, such disagreement does not demonstrate CU’s purposeful avoidance of critical facts.
While we agree with CU, however, that much of Suzuki’s purposeful-avoidance argument boils down to its disapproval of CU’s conclusions, there is one issue that nevertheless precludes summary judgment here. In particular, CU has done nothing to respond to the criticism of its testing procedures as overly influenced by driver input. This evidence formed the basis for the district court’s decision in Isuzu Motors, in which the court relied heavily on the NHTSA report to deny CU’s summary judgment motion, stating that “CU was aware that its tests were significantly reliant upon driver input and skill.” Isuzu Motors, 66 F.Supp.2d at 1125. Suzuki has pointed to further evidence that some CU *1139personnel shared this assessment. The issue is whether CU, armed with the knowledge that its tests were potentially flawed in this way, failed reasonably to investigate in such a manner that could lead a jury to conclude by clear and convincing evidence that CU was aware of the falsity of its Samurai report. Drawing all reasonable inferences in favor of Suzuki on this point, CU’s failure to address this deficiency with its testing procedure could lead a jury to conclude that it was aware that doing so would disclose the falsity of its negative Samurai rating.
Therefore, we conclude that Suzuki has also raised a genuine issue of material fact as to whether CU purposefully avoided information that would have undermined its assessment of the Samurai’s rollover propensity.
IV. CONCLUSION
For the foregoing reasons, the district court’s grant of summary judgment is reversed and the case remanded for further proceedings.15
REVERSED and REMANDED.
Judge Rawlinson was recused.
. For discussion of the need for independent appellate review in the context of copyright, see Eugene Volokh & Brett McDonnell, Freedom of Speech and Independent Judgment Review in Copyright Cases, 107 Yale L.J. 2431 (1998).
. Independent examination is, of course, a one-way street: If the jury rejects a plaintiff's *1114claim, a judge may not invoke the rule to resurrect it.
. The majority's confusion may stem from the fact that Bose and Hoffman were appeals from bench trials. Where the district court is the factfinder, de novo review of its opinion is also de novo review in the New York Times sense because it's effectively de novo review of the underlying question of actual malice. But where the district court merely makes a sufficiency determination- — whether on summary judgment or post-verdict — de novo review of its opinion is not de novo review in the New York Times sense. We always review sufficiency determinations de novo; independent examination means doing something more.
. Judge Graber is of course correct that "[t]he evidence presented at trial often differs markedly from that which is offered in a party’s summary judgment papers.” Concurrence at 1140. That a plaintiff defeats a summary judgment motion doesn't guarantee that the evidence at trial will support a verdict in his favor. What's novel about the majority's approach is that it reverses the rule: It lets a plaintiff defeat summary judgment with a record inadequate to support a verdict, on the mere speculation that some evidence at trial — as yet unknown — will materialize to support his case.
. I use the term "majority” loosely because Judge Graber never tells us which parts of Judge Tashima’s opinion she relies on to find actual malice.
. The introductory explanation to the article is reprinted in full in the appendix.
. The majority also relies on the fact that the driver who made the Samurai tip the first time was not one of CU’s regulars. This fact was also disclosed in CU’s article, see p. 1116 supra, and, in any case, it's hard to see why it matters. Presumably, consumers driving the Samurai are typically not accredited test drivers either. Moreover, it's undisputed that the actual short course test runs, in which the Samurai tipped several times and the other vehicles didn't tip once, were all performed by CU test drivers.
. Even if Judge Graber were right and CU somehow misled the public about its reasons for building the short course, I can't see how it could possibly matter. Whether CU built the short course for the reasons it gave or because little green men from Mars told it to do so, the fact remains that it built the course and the Samurai flunked it while every other vehicle passed. CU’s motives simply are not germane to Suzuki's theory of recovery because they do not in any way defame the Samurai or contribute to any damage it may have suffered. It's as if CU claimed it conducted the tests on a Wednesday when it actually conducted them on a Friday.
. As Suzuki points out, around this time CU made a significant financial outlay in order to secure a new headquarters building in Yonkers, New York. This outlay of approximately $30 million placed CU in what the district court termed a "financially overextended” position.
. The long-course maneuvers followed preliminary testing on the Samurai. CU notes that during the evaluation and break-in process, Alan Hanks, the Facilities Manager of the ATD, rolled the Samurai on a snow-covered gravel road at 15 miles per hour. During the formal test ride on April 7, 1988, CU driver Kevin Sheehan reported that the Samurai was "by far the worst ride in my 20 + years at CU.” Sheehan also took the Samurai on a one-day trip test on April 11, 1988, after which he noted that the Samurai should be rated "Not Acceptable.” In addition, as Suzuki notes, preliminary research had been conducted for the Samurai story prior to the initiation of long-course testing.
. Suzuki contends that the outriggers allowed CU to achieve dramatic results during the tests without risking a real rollover.
. Suzuki emphasizes CU’s submission of what it characterizes as a "false” affidavit in connection with Nappi's statement. CU employees Hanks, Sheehan, Wood, Nappi, and Knoll submitted an affidavit stating that it was De-nison, not Nappi, who could be heard on the videotape saying "All right, Ricky baby” after Small tipped the Samurai on April 26, 1988. Nappi later admitted, however, that he in fact made the statement. Suzuki suggests that the affidavit was submitted in an effort to discredit Denison, who had testified that Landau threatened to find someone to roll the Samurai after the April 20, 1988, long-course testing. CU proffers a benign explanation for Nappi’s changing his story. Resolution of this dispute, however, is not critical to our decision.
. Suzuki also argues that CU deliberately excised footage of the Isuzu Trooper II’s runs from the videotape screened at the press conference to downplay the fact that the Trooper also had trouble making it through the course.
. On September 25, 1996, NHTSA denied a second Samurai defect petition on similar grounds. Denial of Motor Vehicle Defect Petition, 61 Fed.Reg. 50,372 (Sept. 25, 1996).
. In its summary of the test results on multipurpose vehicles, Britain's Department of Transport similarly concluded that the "results from the modified Consumer Union tests were unpredictable" and that "[d]river influence is greatest in the modified Consumer Union manoeuvre.” Dep't of Transp., Stability of Multi-Purpose 4-Wheel Drive Vehicles (Dec. 16, 1988).
. In November 1995, CU published a car buyers guide on CD-Rom that also reiterated the "Not Acceptable” rating from its July 1988 Consumer Reports article.
. After the action was filed, CU continued publicly to refer to the Samurai test, citing the “Not Acceptable” rating in a June 1996 Consumer Reports article on SUVs; sending out contribution solicitation cards in August and October 1996 with a photograph showing the Samurai tipped up on two wheels; and referring to the Samurai in a 1996 press conference and article about the Isuzu Trooper.
. The cases cited by the dissent do not require otherwise. See Andersen v. McCotter, 100 F.3d 723, 725 (10th Cir.1996) (affirming that "[s]ummary judgment is appropriate when there is no genuine issue as to any material fact”); Secrist v. Harkin, 874 F.2d 1244, 1251 (8th Cir.1989) ("To withstand a motion for summary judgment, a public official or public figure must present evidence to support a jury finding that he or she has shown with convincing clarity that a defendant acted with actual malice.”); Liberty Lobby, Inc. v. Dow Jones & Co., 838 F.2d 1287, 1293 (D.C.Cir.1988) ("The question for the court is ‘whether the evidence presented is such that a reasonable jury might find that actual malice had been shown with convincing clarity.’ ” (quoting Anderson, 477 U.S. at 257, 106 S.Ct. 2505)); Herbert v. Lando, 781 F.2d 298, 305 (2d Cir.1986) (affirming summary judgment standards).
Similarly, CU also misstates the applicable standard of review, contending that the normal summary judgment standard does not apply. The cases on which it relies, however, all involve the review of a judgment rendered after trial. See Bose, 466 U.S. at 510-11, 104 S.Ct. 1949; Newton v. Nat’l Broad. Co., 930 F.2d 662, 669 (9th Cir.1991); Eastwood, 123 F.3d at 1252. Thus, these cases do not, as CU suggests, require that we abandon the established standard of review in the summary judgment context.
. Suzuki accepts the actual malice formulation of the product disparagement standard for the purposes of this appeal, although it suggests that the Supreme Court has left this issue open, see Bose, 466 U.S. at 513, 104 S.Ct. 1949, and reserves the right to challenge the standard on further review. Amicus Washington Legal Foundation devotes a substantial portion of its brief to arguing that the First Amendment does not demand a showing of actual malice for product disparagement claims. We decline, however, to address an issue raised only by an amicus. See Russian River Watershed Prot. Comm. v. City of Santa Rosa, 142 F.3d 1136, 1141 n. 1 (9th Cir.1998) ("Generally, we will not consider on appeal an issue raised only by an amicus.”).
. The record does not indicate whether 10 runs was predetermined by the testing protocol or whether Pittle simply stopped making further runs after achieving a tip-up.
. Suzuki also advances a number of additional arguments in support of its contention that the summary judgment record is sufficient to show that CU acted with a high degree of awareness of the probable falsity of its report on the Samurai. Specifically, Suzuki contends that a jury could draw the inference that CU acted with actual malice based on the factual inaccuracies in the Samurai story; CU's premeditation in publicizing false statements about the Samurai; and CU’s concealment of evidence contrary to its claim that the Samurai "rolls over too easily.” Because the analysis set forth above is dispos-itive of the first prong of the Harte-Hanlcs test ("a high degree of awareness of probable falsity”), we need not reach Suzuki's remaining arguments in support of reversal based on Harte-Hanks ' first prong.
. Relying on expert witness testimony, Suzuki asserts that CU violated accepted journalistic standards in failing to engage in further investigation of contradictory evidence. CU responds that the expert testimony is irrelevant because it is not probative of CU’s subjective state of mind. See, e.g., Harris v. Quadracci, 856 F.Supp. 513, 519 (E.D.Wis.1994). Although expert testimony regarding CU’s departure from accepted professional standards is not sufficient by itself to establish actual malice, see Harte-Hanks, 491 U.S. at 669, 109 S.Ct. 2678, it does shed light on the propriety of CU's response to contrary rollover evidence and, thus, is entitled to be given appropriate weight.
. We, of course, intimate no view of the ultimate outcome on the merits.