Schering Corporation v. Pfizer Inc. And Ucb Pharma, Inc.

SOTOMAYOR, Circuit Judge: This appeal invites us to revisit an increasingly important issue in the law of evidence, and one that has confused many courts. Today we clarify the circumstances under which scientifically conducted surveys may be admitted into evidence over a hearsay objection. The case comes to us on appeal from an order of the United States District Court for the Southern District of New York (McKenna, J.) denying a motion for a preliminary injunction made by plaintiff-appellant Schering Corporation (“Schering”) against defendants-appellees Pfizer Inc. (“Pfizer”) and UCB Pharma, Inc. (“UCB”). The primary evidence Schering offered in support of its motion were the results of five surveys in which physicians were asked to relate their memories, and sometimes their impressions, of communications that defendants’ sales representatives had made in the course of their promotional campaigns. Schering offered these surveys to support the more general proposition that defendants’ representatives were engaging in widespread false promotional activity in violation of both the Lanham Trade-Mark Act, 15 U.S.C. § 1051 et seq. (1994) (the “Lanham Act”), and a settlement agreement between the parties. The district court excluded all five surveys on hearsay grounds and then ruled that there was insufficient evidence to support a preliminary injunction. We hold that two of the surveys in this case, which polled physicians’ impressions of the communications at issue, should have been admitted under the present state of mind exception to the hearsay rule, see FedR.Evid. 803(3), for the limited purpose of establishing a pattern of implied falsehood. We also find that in determining whether to admit the five surveys to establish literal falsehoods under the residual hearsay rule, see Fed.R.Evid. 807, the district court relied on an erroneous per se rulé against memory surveys offered to prove the facts remembered. The court thereby abused its discretion, and we remand to the court to determine the surveys’ trustworthiness on the basis of their methodological strengths and their relative susceptibilities to the risks of faulty memory and perception. Finally, we hold that the district court should have admitted one of the surveys, which Pfizer itself commissioned, as well as Pfizer’s internal analysis of this survey as party admissions. See Fed.R.Evid. 801(d)(2). Because the survey evidence was integral to Schering’s case, we vacate the district court’s denial of a preliminary injunction and remand for reconsideration. BACKGROUND Schering is a pharmaceutical corporation that produces Claritin, a leading prescription antihistamine. A factor that strongly contributes to success in the antihistamine market is whether a drug can deliver effective relief from hay fever and other allergy symptoms without causing drowsiness. The first generation of antihistamines, introduced in the 1940s, caused drowsiness in a high percentage of users, and modern over-the-counter antihistamines have this same effect. Claritin, by contrast, is a new, second-generation prescription antihistamine, launched in 1993, which causes no more sedation than placebos in clinical tests. This feature has been important to Claritin’s market success. UCB is a European pharmaceutical company that has developed a competing second-generation prescription antihistamine called “Zyrtec.” Because of UCB’s limited *222 presence in the United States, UCB licensed Pfizer, a Delaware corporation, to co-proraote the product domestically. Regulations promulgated by the United States Food and Drug Administration (“FDA”) required Pfizer to perform several controlled clinical tests on Zyrtec before the drug could be registered for domestic use, however, and these tests revealed that Zyrtec causes approximately twice as much sedation as a placebo. The FDA therefore required Pfizer to caution both physicians and consumers, through proper labeling and warning instructions, that Zyrtec has these sedating qualities. The FDA also warned Pfizer that it would be misleading to advertise Zyrtec by focusing on the rate at which customers discontinued using the drug because of somnolence (ie., 1%) rather than the drug’s somnolence level itself (ie., 11-14%). In early 1996, Pfizer and UCB began selling Zyrtec in the United States. To promote the product, Pfizer used a method that is common in the pharmaceutical industry: it employed a team of approximately 1200 sales representatives to visit physicians across the nation and emphasize the product’s qualities in one-on-one informational meetings called “detailings.” In these meetings, Pfizer representatives were to promote Zyrtec, persuade doctors to prescribe it more often and respond to any questions or concerns that doctors might have about the drug. No records were kept of these meetings. Suspicious that Zyrtec representatives might be misrepresenting the product’s somnolence levels in the detailings, Scher-ing hired IMS America, a commercial vendor of market surveys, to conduct a survey of physicians to determine whether any such misrepresentations were being made. The survey results led Schering to believe that they were. In February 1996, Scher-ing thus filed suit against Pfizer for violating Section 43(a)(2) of the Lanham Act, 15 U.S.C. § 1125(a)(2) (1994), by false advertising. See Schering Corp. v. Pfizer, Inc., No. 96 Civ. 1462(LMM) (S.D.N.Y. Feb. 29, 1996). On April 4, 1996, this dispute ended in a settlement agreement (the “Settlement Agreement”), under which Pfizer and UCB agreed not to permit their sales representatives to state either ‘expressly’ that Zyrtec was ‘lowsedating’ or ‘expressly or by implication’ that Zyrtec was ‘nonsedat-ing’ or ‘essentially nonsedating.’ These restrictions applied to all forms of advertising, including “verbal statements made to doctors.” In order to monitor Pfizer’s and UCB’s compliance with the Settlement Agreement, Schering hired DTW in early 1998 to perform another survey of physicians concerning statements being made in the detailings. This survey polled 78 physicians who had been detailed between March 30 and June 5, 1998, and who were allegedly representative of a panel of over 6000 physicians nationwide who prescribe high volumes of antihistamines. 1 The doctors were asked the following question on the same day as the detailings: What did the sales representatives tell you about [Zyrtec]? Please be as specific and complete as you can in describing the message or information that was conveyed to you about [Zyrtec]. The survey results suggested that approximately 30% of the Zyrtec agents were representing that the drug was either low sedating or nonsedating. Schering subsequently commissioned a second survey from DTW, which was customized to be representative of a panel of over 6000 physicians nationwide who prescribe antihistamines in any amount. The survey polled 98 physicians shortly after they were detailed and asked more specifically: “What did the representative say about the product and sedation?” Approximately half of the physicians polled re *223 ported a representative describing Zyrtec as low sedating or nonsedating. Sehering subsequently wrote a letter to Pfizer dated June 19, 1998, complaining of pervasive violations of the Settlement Agreement based on the results of these surveys. Pfizer denied the violations but assured Sehering that it would remind its sales force to comply with the mandates of the agreement. Sehering then continued to monitor Pfizer’s compliance by commissioning another survey, this time from the firm Clarke, Matire and Bartolomeo (“CMB”). This survey polled 200 physicians, who were selected randomly from those who prescribe antihistamines in above-minimal quantities nationwide. The interviews took place within a day of the detailings, between July 23 and September 17, 1998, and included the following question: “During (today’s/yesterday’s) Zyrtec detail, did the rep say anything about Zyrtec tablets and sedation?” The survey indicated that 47% of Zyrtec representatives were calling the product either low sedating or nonsedating. On October 5, 1998, Sehering brought suit in the United States District Court for the Southern District of New York seeking a preliminary injunction against alleged ongoing and pervasive breaches of the Settlement Agreement and Section 43(a)(2) of the Lanham Act. In the course of discovery, Sehering learned that soon after receiving Schering’s original letter of complaint, Pfizer had commissioned Market Measures Inc. (“MMI”) to perform a “Fas-Tape Survey,” a common type of market survey that probes the main messages conveyed in an advertising campaign, “to understand the key messages and competitive claims concerning the sedation side effect pertaining to Zyrtec.” This survey involved a random sample of 74 physicians, who were allegedly representative of a panel of over 20,000 doctors prescribing antihistamines nationwide. The physicians were detailed between July 7 and July 12, 1998, and were instructed to call an 800 number immediately after receiving a sales presentation. They were then asked: “In one or two sentences, what was the main message of the presentation?” The FasTape Survey stated that “Zyrtec representatives appear to be focusing almost equally on the drug’s indications ... and efficacy, folloioed by its loio/no sedation.” (Emphasis added). During discovery, Sehering also uncovered an analysis of this FasTape Survey, prepared by Lakshmi Gengler, a market research executive at Pfizer. This internal analysis stated that “physicians report one-third of Zyrtec sales representatives mentioning low/non sedation with respect to Zyrtec’s side effect profile and mechanism of action.” During discovery, Sehering also learned that Pfizer had commissioned Seott-Levin, another well-known market research company, to conduct a survey to identify the main messages being conveyed in the de-tailings. This survey, unlike the others, asked only about messages concerning nonsedation and did not poll for messages concerning low sedation. Sehering obtained several pages of this survey, which indicated that Zyrtec representatives were suggesting at a rate of approximately 21% in June of 1998 and 15% in July of 1998 that Zyrtec was nonsedating. Because Sehering was unable to obtain the entire survey, Sehering commissioned its own Seott-Levin Survey, which polled approximately 150 physicians per month from March through August of 1998. The survey results suggested that on average, approximately 16% of the Zyrtec agents were communicating a message that Zyrtec was nonsedating. At the hearing on the preliminary injunction motion, Sehering sought to introduce five surveys: the two DTW surveys, the CMB Survey, the FasTape Survey and the Scotb-Levin Survey. Sehering also sought to introduce Pfizer’s internal analysis of the FasTape Survey. Finally, Sehering called to the stand Dr. Bartolo-meo, an executive at CMB, who testified that it is “very rare” for so many surveys of these kinds to have such consistent re- *224 suits. Pfizer responded with a motion arguing that all of this evidence contained hearsay and was thus inadmissible under Rule 802 of the Federal Rules of Evidence. The district court agreed and issued a written opinion refusing to admit Scher-ing’s surveys for any purpose. See Schering Corp. v. Pfizer, Inc., No. 98 Civ. 7000(LMM), 1999 WL 144921 (S.D.N.Y. Mar.16, 1999). The only other significant evidence that Schering offered in support of its motion was a series of manuals used by Pfizer to train its sales representatives. These manuals contained some instructions to call Zyrtec “low sedating” and to focus attention away from the drug’s somnolence level and toward the rate at which customers discontinued using the drug because of somnolence effects. The manuals also indicated that eetirizine, Zyrtec’s active ingredient, does “not cause sedation.” The district court found this evidence insufficient to warrant a preliminary injunction and denied Schering’s motion. This appeal followed. DISCUSSION I. Standards of Review We review a district court’s evi-dentiary rulings for abuse of discretion. See, e.g., Phoenix Assocs. III v. Stone, 60 F.3d 95, 100 (2d Cir.1995). “Either an error of law or a clear error of fact may constitute an abuse of discretion.” Charette v. Town of Oyster Bay, 159 F.3d 749, 755 (2d Cir.1998); see also Pullman-Standard v. Swint, 456 U.S. 273, 287, 102 S.Ct. 1781, 72 L.Ed.2d 66 (1982) (holding that a factual finding may be set aside if it “rest[s] on an erroneous view of the law”). An evidentiary ruling that is an abuse of discretion is, however, only reversible if it also affects a party’s substantial rights. See Fed.R.Evid. 103(a). This occurs when, for example, a district court excludes a party’s primary evidence in support of a material fact, and failure to prove that fact defeats the party’s claim. See O’Neal v. Esty, 637 F.2d 846, 848 (2d Cir.1980) Here, there is no dispute that the district court excluded Schering’s primary evidence in support of its motion for a preliminary injunction, and there is no dispute that Schering’s motion was denied for lack of evidence that the surveys may have cured. In particular, the surveys may have helped establish that Pfizer and UCB were engaging in ongoing non-trivial violations of the Settlement Agreement and Section 43(a)(2) of the Lanham Act, 15 U.S.C. 1125(a)(2) (1994), and that a preliminary injunction was thus warranted. In what follows, we therefore limit our inquiry to whether the district court’s exclusion of these surveys was an abuse of discretion. II. Admissibility of Surveys in General In the first half of this century, surveys were generally regarded as inherently untrustworthy because they contained hearsay, or out-of-court statements offered to prove the truth of the matters asserted. See, e.g., DuPont Cellophane Co. v. Waxed Prods., 6 F.Supp. 859, 884 (E.D.N.Y.1934), modified, 85 F.2d 75 (2d Cir.1936) (refusing to admit survey because court could not “see how plaintiff could even test the facts, as it had no opportunity for cross-examination of those who were supposed to have answered the questions”); Elgin Nat’l Watch Co. v. Elgin Clock Co., 26 F.2d 376, 376-77 (D.Del.1928) (refusing to admit survey, even as basis for expert opinion, because survey collected hearsay). Schering argues, however, that “the modern view is that the hearsay objection is without merit and that any technical deficiencies in survey methodology go to [a survey’s] weight as evidence, not to its admissibility.” (Pl.’s Br. at 33-34 (quoting 5 J. Thomas McCarthy, McCarthy on Trademarks and Unfair Competition § 32:169, at 32-248 (4th ed.1998) [hereinafter McCarthy on Trademarks ]).) Under the “modern view,” according to Schering, surveys should be admitted as a general *225 rule, and their weight should, be determined by whether: (1) the “universe” was properly defined, (2) a representative sample of that universe was selected, (3) the questions to be asked of interviewees were framed in a clear, precise and non-leading manner, (4) sound interview procedures were followed by competent interviewers who had no knowledge of the litigation or the purpose for which the survey was conducted, (5) the data gathered was accurately reported, (6) the data was analyzed in accordance with accepted statistical principles and (7) the objectivity of the entire process was ensured. Toys “R” Us, Inc. v. Canarsie Kiddie Shop, Inc., 559 F.Supp. 1189, 1205 (E.D.N.Y.1983) (paraphrasing Manual for Complex Litigation § 2.712 (5th ed.1981)); Hutchinson v. Essence Communications, Inc., 769 F.Supp. 541, 557 (S.D.N.Y.1991); see also Manual for Complex Litigation, Third § 21.493 (1995). These factors derive from accepted principles of survey methodology and help define when a survey has been properly conducted. See, e.g., McCarthy on Trademarks § 32:181, at 32-272. There is no doubt that beginning with Judge Wilfred Feinberg’s seminal decision in Zippo Manufacturing Co. v. Rogers Imports, Inc., 216 F.Supp. 670 (S.D.N.Y.1963), the general trend has been toward the admission of surveys of various kinds. Surveys are, for example, routinely admitted in trademark and false advertising cases to show actual confusion, genericness of a name or secondary meaning, all of which depend on establishing that certain associations have been drawn in the public mind. See, e.g., Bristol-Myers Squibb Co. v. McNeil-P.P.C., Inc., 973 F.2d 1033, 1043 (2d Cir.1992) (secondary meaning); PPX Enters., Inc. v. Audiofidelity Enters., Inc., 818 F.2d 266, 271 (2d Cir.1987) (actual confusion); Nestle Co. v. Chester’s Market, Inc., 571 F.Supp. 763, 769-70, 773-75 (D.Conn.1983) (genericness of name). See generally Marcia B. Paul & Anthony F. Lo Cicero, Litigating Trademark, Section 43(a) and Unfair Competition Cases, P.L.I. Patents, Copyrights, Trademarks, and Literary Property Course Handbook Series No. G4-3925 (Oct.1994) (describing use of surveys in trademark and false advertising context). Surveys of other types have occasionally been admitted for other limited purposes. See, e.g., Keith v. Volpe, 858 F.2d 467, 479-81 (9th Cir.1988) (admitting survey to show statistics concerning respondents’ race, income and housing preferences); Debra P. v. Turlington, 730 F.2d 1405, 1408, 1412-14 (11th Cir.1984) (admitting survey to show “whether the teacherfs] [surveyed] had provided instruction during 1981-82 relating to the skills tested on the SSAT-II and if so, whether that instruction had been sufficient for a student to master the skills”). The use of surveys is, however, far more prevalent in trademark law than in most other areas. See Neal Miller, Facts, Expert Facts, and Statistics, 40 Rutgers L.Rev. 101, 137 (1987). Moreover, contrary to Schering’s contention, the case law does not support a general rule allowing surveys into evidence for all purposes. See, e.g., United States v. Pryba, 900 F.2d 748, 757 (4th Cir.1990) (excluding survey in obscenity case when offered “to demonstrate the community’s attitude, toleration and standards with regard to sexually explicit materials”); American Footwear Corp. v. General Footwear Co., 609 F.2d 655, 660 n. 4, 663 (2d Cir.1979) (holding that district court’s rejection of survey offered to show secondary meaning was not clearly erroneous); Ortho Pharm. Corp. v. Cosprophar, Inc., 828 F.Supp. 1114, 1122 (S.D.N.Y.1993) (excluding survey offered to show actual confusion). A review of the case law suggests that there are, in fact, two ongoing controversies that tend to complicate the question of whether survey evidence should be admitted in a particular case. First, there is a dispute over the proper consequence of a finding of methodological error in a survey. While some courts in this Circuit *226 believe such flaws are proper grounds for exclusion, others view methodological errors as affecting only the weight of the evidence. Compare, e.g., Mobil Oil Corp. v. Pegasus Petroleum Corp., 818 F.2d 254 (2d Cir.1987) (“The district court properly admitted these surveys into evidence, despite claims of statistical imperfections by both sides, as those criticisms affected the weight accorded to the evidence rather than its admissibility.”), Bristol-Myers, 973 F.2d at 1043 (affirming district court’s decision to admit survey and allow methodological errors to affect the weight of the evidence), and Centaur Communications, Ltd. v. A/S/M Communications, Inc., 652 F.Supp. 1105, 1110 (S.D.N.Y.1987) (admitting survey and allowing methodological errors to affect only the weight of the evidence), with, e.g., Ortho Pharm., 828 F.Supp. at 1122 (“This court cannot conclude that the surveys were ‘properly conducted’ ... and therefore concludes ... that the surveys are inadmissible hearsay.”), and Toys “R” Us, 559 F.Supp. at 1205 (excluding survey because of methodological errors). Indeed, this same division runs among, and sometimes within, other circuits. Compare, e.g., Indianapolis Colts, Inc. v. Metropolitan Baltimore Football Club Ltd. Partnership, 34 F.3d 410, 415-16 (7th Cir.1994) (finding that court was correct to draw inferences from survey, despite methodological problems), Brunswick Corp. v. Spinit Reel Co., 832 F.2d 513, 523 (10th Cir.1987) (“As to the technical and methodological deficiencies in the survey that Spinit charges, those relate not the survey’s admissibility but to the weight to be given such evidence.”), Levi Strauss & Co. v. Blue Bell, Inc., 778 F.2d 1352, 1358 (9th Cir.1985) (en banc) (affirming district court’s decision to admit surveys and determine their weight by examining methodology), Zatarains, Inc. v. Oak Grove Smokehouse, Inc., 698 F.2d 786, 795, 797 (5th Cir.1983) (finding that “survey evidence is the most direct and persuasive way to establish secondary meaning,” but that particular survey at issue was entitled to “little evidentiary weight” because of its methodological defects), and President and Trustees of Colby College v. Colby College-New Hampshire, 508 F.2d 804, 809 (1st Cir.1975) (finding it clearly erroneous to reject survey altogether, and giving survey some weight, despite methodological defects), with, e.g., Harolds Stores, Inc. v. Dillard Dep’t Stores, Inc., 82 F.3d 1533, 1544 (10th Cir.1996) (“We allow the admission of survey evidence as an exception to the hearsay rule if the survey is material, more probative on the issue than other evidence and if it has guarantees of trustworthiness.” (citations omitted) (internal quotations omitted)), C.A. May Marine Supply Co. v. Brunswick Corp., 649 F.2d 1049, 1054 (5th Cir.1981) (“Surveys and customer questionnaires are admissible, if they are pertinent to the inquiry, upon a showing that the poll is reliable and was compiled in accordance with accepted survey methods.”), Baumholser v. Amax Coal Co., 630 F.2d 550, 552 (7th Cir.1980) (“To qualify a study or opinion poll for admission into evidence, there must be a substantial showing of reliability.”), and Pittsburgh Press Club v. United States, 579 F.2d 751, 758 (3d Cir.1978) (holding that surveys are admissible only if they are “conducted with proper safeguards to insure accuracy and reliability”). Courts have, however, rarely articulated their reasons for falling onto one or another side of this divide. The second dispute reflected in the case law involves the proper rationale for allowing admissible surveys into evidence. As Judge Feinberg explained in Zippo: Some cases hold that surveys are not hearsay at all; other cases hold that surveys are hearsay but are admissible because they are within the recognized exception to the hearsay rule for statements of present state of mind, attitude, or belief. 216 F.Supp. at 682; see also Grotrian, Helfferich, Schulz, Th. Steinweg Nachf. v. Steinway & Sons, 523 F.2d 1331, 1341 (2d *227 Cir.1975) (‘We are mindful of the on-going dispute as to whether a survey should be admissible ... under the state of mind exception to the hearsay rule or based on the need for it plus adequate guarantees of trustworthiness.”). We have sometimes suggested that surveys are properly admitted under the residual hearsay rule, see, e.g., Starter Corp. v. Converse, Inc., 170 F.3d 286, 297 (2d Cir.1999) (citing residual hearsay rule as basis for admitting survey), and the Advisory Committee for the Federal Rules of Evidence has suggested that surveys are sometimes best admitted as bases of expert testimony pursuant to Rule 703, see, e.g., Fed.R.Evid. 703 advisory committee’s note (“[Rule 703] offers a more satisfactory basis for ruling upon the admissibility of public opinion poll evidence. Attention is directed to the validity of the techniques employed rather than to relatively fruitless inquiries whether hearsay is involved.”). Quite often, courts simply fail to cite a statutory basis for an admission. See Zippo, 216 F.Supp. at 682. These confusions in the case law have led at least one prominent commentator to conclude that [a] skeptic would classify the survey cases into two categories: a survey is accepted and relied upon when the judge already has his or her mind made up in favor of the survey results; and a survey is rejected and torn apart when the judge subjectively disagrees with the survey results. McCarthy on Trademarks § 32:196, at 32-298. This skepticism is unwarranted. Because the Federal Rules of Evidence govern the admissibility of all evidence over a hearsay objection, the question of survey admissibility is ultimately a question of statutory interpretation. As shown in the following two sections, careful attention to the possible statutory grounds 'for admitting hearsay can help harmonize most of the case law. Whether the district court was correct to exclude the five surveys in this case thus merits detailed discussion, with particular attention to the types of statements contained in the surveys, the purposes for which they were offered and the various possible grounds for their admission. III. The State of Mind Exception: Admissibility to Show What Was Implied One of the two most common bases for admitting survey evidence is Rule 803(3), which creates an exception to the hearsay rule for statements that express a declarant’s state of mind at the time of the utterance. In particular, Rule 803(3) excepts any statement of [a] declarant’s then existing state of mind, emotion, sensation, or physical condition (such as intent, plan, motive, design, mental feeling, pain, and bodily health), but not including a statement of memory or belief to prove the fact remembered or believed unless it relates to the execution, revocation, identification, or terms of declarant’s will. Fed.R.Evid. 803(3). The great majority of surveys admitted in this Circuit, including those used in Lanham Act cases to establish actual confusion or secondary meaning, fall into this category: they poll individuals about their presently-existing states of mind to establish facts about the group’s mental impressions. See, e.g., Sterling Drug, Inc. v. Bayer AG, 14 F.3d 733, 741-42 (2d Cir.1994) (polling consumers for their then-existing perceptions to establish actual confusion); Mobil Oil Corp., 818 F.2d at 259 (same); Harlequin Enters. Ltd. v. Gulf & Western Corp., 644 F.2d 946, 949-50 (2d Cir.1981) (polling consumers for their then-existing states of mind to establish secondary meaning, or tendencies to associate certain product features with a particular corporate origin). It is important for district courts to recognize surveys of this type because their qualification for a traditional hearsay exception obviates the need to examine methodology before overruling a hearsay *228 objection. Regardless of the basis cited for admitting these surveys, errors in methodology thus properly go only to the weight of the evidence—subject, of course, to Rule 403’s more general prohibition against evidence that is less probative than prejudicial or confusing. See, e.g., Grotrian, 523 F.2d at 1341 (finding, in the context of a survey offered to show actual confusion, that while courts cite different bases for admitting survey evidence, “[t]he more relevant issue [is] the weight to be given to the surveys”); see also Starter, 170 F.3d at 297 (holding that survey evidence offered to show actual confusion was properly excludable under Rule 403 when it was so flawed that its probative value was outweighed by the risk of prejudice or confusion). The “modern view” urged by Schering is thus fully applicable to these kinds of surveys. See McCarthy on Trademarks § 32:168, at 32-246 (“[S]ince at least 1951 the cases are now unanimous that evidence of the state of mind of persons surveyed is not inadmissible as hearsay.” (footnote omitted)) (collecting cases). See generally Jack P. Lipton, Trademark Litigation: A New Look at the Use of Social Science Evidence in Trademark Litigation, 29 Ariz. L.Rev. 639 (1987). But cf. Section IV(i), infra (explaining that where residual hearsay rule is only possible basis for admission, methodological defects must be examined to determine not only weight but admissibility of survey evidence). Schering argues that both the Fas-Tape and Scott Levin surveys fall into this category because they asked physicians to relate thé “main messages” conveyed by Zyrtee agents in the detailings. The surveys thereby asked physicians to relate not only what was said in the meetings but the impressions with which they were left. Such impressions are classic states of mind and, as such, fall under Rule 803(3). See, e.g., Zippo, 216 F.Supp. at 683-84; Telebrands Corp. v. E. Mishan & Sons, No. 97 Civ. 1414(RPP), 1997 WL 232595, at *20 (S.D.N.Y. May 7, 1997). The district court did not address directly whether these surveys polled for then-existing states of mind. Instead, the court appears to have viewed evidence of these states as irrelevant to the present litigation. The court explained: The Court does not fully concur with Schering’s suggestion that the Court need not conclude that the surveys report “what exact words were spoken.” While it may not need to know the exact words spoken, the Court must know the exact substance of what was said by the representatives to the physicians, because it is that substance that violates, or does not violate, the settlement agreement. To determine that substance requires an effort to determine, as closely as possible, the exact words that were spoken. Schering Corp., 1999 WL 144921, at *4 n. 4 (emphasis added) (citation omitted). We disagree. The Settlement Agreement states in relevant part that: Pfizer and UCB Pharma hereby agree that, in connection with their advertising and promotion of ZYRTEC in the United States, they will not claim or allow those acting on their behalf to claim, either expressly or by implication that: (1) ZYRTEC and/or its active ingredient cetirizine is nonsedating or essentially nonsedating; (2) ZYRTEC and/or cetirizine is as non-sedating as CLARITIN .... or is comparable to CLARITIN ... or to the “second” generation of antihistamines ... in terms of its somnolence or sedation; (3) ZYRTEC’s and/or cetirizine’s discontinuance rate due to somnolence means that ZYRTEC and/or cetirizine is non-sedating or essentially non-sedating; (4) After a short period of time, ZYRTEC and/or cetirizine users *229 who experience sedation develop tolerance to its sedating effect. (Settlement Agreement of 4/4/1996, at 3-4 (emphasis added).) The Settlement Agreement thus prohibits not only explicit but also implicit falsehoods. In this sense, the agreement is reminiscent of the Act upon which it was based — i.e., the Lanham Act — which prohibits advertisements that are not only literally but also impliedly false. See L & F Prods. v. Procter & Gamble Co., 45 F.3d 709, 711 (2d Cir.1995) (noting that Lanham Act prohibits impliedly false advertising, or advertisements that “although literally true, [are] still likely to mislead or confuse consumers”); see also S.C. Johnson & Son, Inc. v. Carter-Wallace, Inc., 614 F.Supp. 1278, 1319 (S.D.N.Y.1985) (“There are two types of actionable false advertising: (1) advertising which makes claims which are literally false on their face, and (2) advertising which, although literally true on its face, is perceived by a significant proportion of the relevant market as making ‘subliminal’ or ‘implicit’ claims which are provably false. With regard to the second type of false advertising, the courts sometimes say that the advertising has a tendency to ‘mislead, confuse or deceive.’ ”). 2 Cases in the Lanham Act context demonstrate, moreover, that the mental impressions with which an audience is left can be relevant, and sometimes even necessary, to establish what a defendant is implying in a challenged representation. See, e.g., Cocar-Cola Co. v. Tropicana Prods., Inc., 690 F.2d 312, 317 (2nd Cir.1982) (“When a merchandising statement or representation is literally or explicitly false, the court may grant relief without reference to the advertisement’s impact on the buying public. When[, however,] the challenged rather than explicitly false, its tendency to violate the Lanham Act by misleading, confusing or deceiving should be tested by public reaction.” (emphasis added) (citations omitted)). In fact, although plaintiffs seeking to establish a literal falsehood must generally show the substance of what is conveyed, we have held that a “district court must rely on extrinsic evidence to support a finding of an implicitly false message.” See Johnson & Johnson * Merck Consumer Pharms. Co. v. Smithkline Beecham Corp., 960 F.2d 294, 297 (2d Cir.1992) (emphasis added). This is because plaintiffs alleging a literal falsehood are claiming that a statement, on its face, conflicts with reality, a claim that is best supported by comparing the statement itself with the reality it purports to describe. By contrast, plaintiffs alleging an implied falsehood are claiming that a statement, whatever its literal truth, has left an impression on the listener that conflicts with reality. This latter claim invites a comparison of the impression, rather than the statement, with the truth. See generally Jacob Jacoby et al, Survey Evidence in Deceptive Advertising Cases Under the Lanham Act: An Historical Review of Comments from the Bench, 954 PLI/Corp. 83, 87-88 (1996). Given this distinction in the false advertising context, we see no reason to believe that Schering’s allegations of breach of the Settlement Agreement by implied falsehood cannot similarly be tested by this type of evidence. Moreover, Sehering rested its argument for a preliminary injunction not only on alleged breaches of the Settlement Agreement but also on alleged violations of the Lanham Act. *230 These considerations persuade us that the FasTape and Scott Levin surveys are relevant to show ongoing violations of both the Settlement Agreement and the Lan-ham Act, on a theory that Zyrtec agents were making statements that regardless of their veracity, left the physicians with false impressions. Although statements by out-of-court declarants—e.g., the surveyed physicians—-relating such impressions are hearsay where, as here, they are offered to establish the existence of the impressions themselves, 3 see Fed.R.Evid. 801(c), these declarations are independently admissible under Rule 803(3) as expressions of the declarant physicians’ then-existing states of mind. The district court should have thus admitted the FastTape and Scott Levin surveys for the limited purpose of establishing a pattern of implied falsehood. This ruling concerning admissibility in no way suggests that the district court should give these surveys any particular weight on remand. The fact that the two surveys corroborate one another is certainly a point in their favor, as is the fact that Schering produced independent evidence of Pfizer training manuals, which appear to have recommended misleading promotional practices. Still, Pfizer has offered a series of criticisms concerning the surveys’ methodologies and the potentially leading or ambiguous nature of the questions they posed. Schering also has failed to bring even a small subset of the surveyed physicians into court, a process that would have allowed for their cross-examination and would have greatly assisted the court in determining what weight to give the surveys overall. Finally, Pfizer has argued that the FDA does not draw a sharp distinction between “sedating” and “nonsedating” antihistamines, and that Schering fabricated this distinction to generate surveys that would indicate falsehoods when there were none. We therefore leave it to the district court to assess the weight of all the evidence admitted after making more detailed factual findings concerning these and related issues. IV. The Residual Hearsay Rule: Admissibility to Show What Was Said The central reason that Schering sought to introduce its five surveys was to establish not that Zyrtec representatives were leaving physicians with false impressions but that the representatives were stating literal falsehoods in violation of the Lanham Act and the Settlement Agreement. Used for this purpose, the surveys can still be characterized as polling then-existing states of mind because they asked the physicians to relate memories of statements made in the detailings, or beliefs about what the sales representatives had *231 said, both of which are technically states of mind. If statements expressing such states are offered to establish not only that the states existed, however, but also the facts thereby recalled or believed, the statements are no longer simply expressions of mental state. In fact, Rule 803(8) explicitly excludes from its purview any “statement of memory or belief to prove the fact[s] remembered or believed.” Fed. R.Evid. 803(3). The district court thus correctly refused to admit Schering’s surveys under the present state-of-mind exception to establish what Zyrtec agents were literally saying in their promotional activities. Schering has, however, cited a second common basis for admitting the survey for this purpose: the residual hearsay rule. This rule, presently codified at Rule 807 of the Federal Rules of Evidence, states that: A statement not specifically covered by Rule 803 or 804 but having equivalent circumstantial guarantees of trustworthiness, is not excluded by the hearsay rule, if the court determines that (A) the statement is offered as evidence of a material fact; (B) the statement is more probative on the point' for which it is offered than any other evidence which the proponent can procure through reasonable efforts; and (C) the general purposes of these rules and the interests of justice will best be served by admission of the statement into evidence. However, a statement may not be admitted under this exception unless the proponent of it makes known to the adverse party sufficiently in advance of the trial or hearing to provide the adverse party with a fair opportunity to prepare to meet it, the proponent’s intention to offer the statement and the particulars of it, including the name and address of the declarant. Fed.R.Evid. 807 (emphasis added). “To be admissible under this exception, ‘the evidence must[, in other words,] fulfill five requirements: trustworthiness, materiality, probative importance, the interests of justice and notice.’ ” United States v. Harwood, 998 F.2d 91, 98 (2d Cir.1993) (quoting Parsons v. Honeywell, Inc., 929 F.2d 901, 907 (2d Cir.1991) (citations omitted)). The district court found Schering’s surveys inadmissible under the residual hearsay rule to show what Zyrtec representatives were literally saying on three independent grounds: (i) the surveys had insufficient circumstantial guarantees of trustworthiness; (ii) they were not more probative to show what was said in the detailings than any other evidence that Schering could procure through reasonable efforts; and (iii) the general purposes of the Federal Rules of Evidence and the interests of justice would not be best served by admitting the surveys. Schering argues that in reaching each of these conclusions, the district court abused its discretion by relying on an erroneous per se rule against the use of out-of-court memory statements to prove the facts remembered. See generally Pullman-Standard, 456 U.S. at 287, 102 S.Ct. 1781 (holding that judgments “restfing] on an erroneous view of the law” may be set aside on that basis); Charette, 159 F.3d at 755 (holding that clear error is species of abuse of discretion). We agree. i. Trustworthiness of the Surveys In deciding that the surveys did not have sufficient guarantees of trustworthiness under the residual hearsay rule to help establish what Zyrtec agents were saying, the district court relied heavily on the same fact that made Rule 803(3) inapplicable for this purpose: the surveys elicited memory statements offered to establish the occurrence of the remembered events. The court explained that [a]fter speaking of the [Residual Hearsay] Rule’s “theory that under appropriate circumstances a hearsay statement may possess circumstantial guarantees of trustworthiness sufficient to justify *232 nonproduction of the declarant in person at the trial even though he may be available,” a theory which “finds vast support in the many exceptions to the hearsay rule ...the [Advisory] Committee went on to say of Rule 803(8)-the closest analogy to the present situation-that “[t]he exclusion of‘statements of memory or belief to prove the fact remembered or believed’ is necessary to avoid the virtual destruction of the hearsay rule which would otherwise result from allowing a state of mind, provable by a hearsay statement, to serve as the basis for an inference of the happening of the event which produced the state of mind.” Sobering, 1999 WL 144921, at *4 (emphasis added) (citations omitted). The court thus drew an explicit analogy between the analyses of memory statements under Rule 803(3) and under the residual hearsay rule. It is, however, incorrect to project this Rule 803(3) limitation into the residual hearsay rule context. Unlike Rule 803(3), which explicitly excludes from its purview memory statements offered to establish the facts remembered, the residual hearsay rule contains no such express limitation. There is a reason for this difference. Almost any statement used to describe events that a speaker has experienced in the past can be characterized as a “memory,” which is a presently-existing state of mind when it is conveyed. If such statements were admissible under Rule 803(3) to prove the facts remembered, parties could thus offer hearsay to establish almost any past fact, a result that would indeed mark “the virtual destruction of the hearsay rule.” Fed.R.Evid. 803(3) advisory committee’s note. The residual hearsay rule, by contrast, escapes this problem by setting forth its own set of requirements, which include necessity and trustworthiness, before it will allow for a statement’s admission. These requirements independently ensure that the rule will “be used very rarely, and only in exceptional circumstances.” S.Rep. No. 93-1277, at 36 (1974), reprinted in 1974 U.S.C.C.A.N. 7051, 7062 (quoted with approval in United States v. Mahler, 579 F.2d 730, 736 (2d Cir.1978)). Moreover, by allowing for the admission of memory surveys under the residual hearsay rule only when they would be independently admissible under Rule 803(3), the district court effectively read the residual hearsay rule as incompetent to perform any independent work in these circumstances. As Judge Feinberg explained in Zippo, however, surveys can be admitted at times even if they do “not fit within th[e] [present state of mind] exception,” based upon “the need for the statement at trial and the circumstantial guaranty of trustworthiness surrounding the making of the statement.” 216 F.Supp. at 683. The district court thus erred by relying on a per se rule against the trustworthiness of memory surveys. A close examination of the function of the hearsay rule will illuminate the criterion of trustworthiness that the district court should have employed. The hearsay rule is generally said to exclude out-of-court statements offered for the truth of the matter asserted because there are four classes of risk peculiar to this kind of evidence: those of (1) insincerity, (2) faulty perception, (3) faulty memory and (4) faulty narration, each of which decreases the reliability of the inference from the statement made to the conclusion for which it is offered. See, e.g., Headley v. Tilghman, 53 F.3d 472, 477 (2d Cir.1995) (referring to classic hearsay “risks of insincerity, distorted perception, imperfect memory, and ambiguity of utterance”); Edmund M. Morgan, Hearsay Dangers and Application of the Hearsay Concept, 62 Harv. L.Rev. 177, 185 (1948) (identifying these classic hearsay risks). The hearsay rule ordinarily prohibits the admission of out-of-court statements by declarants on the theory that cross-examination can help test for these four classes of error, thus allowing the fact-finder to weigh the evidence properly and to discount any that is too unreliable. See Note, Public Opinion Surveys as Evidence: The Pollsters Go to *233 Court, 66 Harv. L.Rev. 498, 501 (1953) (“Cross-examination serves to test the de-clarant’s sincerity, narrative ability, perception, and memory....”). The traditional exceptions to the hearsay rule, in turn, provide the benchmark against which the trustworthiness of evidence must be compared in a residual hearsay analysis. See Fed.R.Evid. 807 (allowing for admission of hearsay statements not specifically covered by Rules 803 or 804 but having, among other things, “equivalent circumstantial guarantees of trustworthiness”). It is thus important to recognize that the trustworthiness of these exceptions is a function of their ability to minimize some of the four classic hearsay dangers. See generally Laurence H. Tribe, Triangulating Hearsay, 87 Harv. L.Rev. 957, 961 (1974) (“The traditional hearsay exceptions can usefully be clustered into*... groups, based on the different ways in which the members of each group overcome the [four] potential infirmities [generally associated with hearsay].”). Statements based on present sense impressions, for example, do not suffer from the risk of faulty memory because they are made at or near the time of impression. These statements also express knowledge based on direct sensory perception. The Federal Rules of Evidence thus make an exception to the hearsay rule for “statements] describing or explaining an event or condition made while the declarant was perceiving the event or condition, or immediately thereafter.” Fed.R.Evid. 803(1). Similarly, statements falling under Rule 803(3)’s exception for presently-existing states of mind rarely suffer from the risks of faulty memory because they are made when the declarant is in the relevant state, and they bear minimal risk of faulty perception because speakers generally know their own states of mind. See Tribe, 87 Harv. L.Rev. at 965 (“The closeness in time of statement to perception reduces memory problems to the de minimis level, and for a number of reasons, including the fact that what one perceives as his physical or mental sensations are his sensations, there is ordinarily no possibility of erroneous perception.”). Although both of these kinds of statement can suffer from the remaining risks of insincerity and faulty narration, exceptions to the hearsay rule are made for them on “the theory that ... [the statements] possess circumstantial guarantees of trustworthiness sufficient to justify nonproduction of the declarant in person at the trial even though he may be available.” Fed.R.Evid. 803 advisory committee’s note. It follows that a hearsay statement need not be free from all four categories of risk to be admitted under Rule 807. Courts deciding whether evidence is sufficiently trustworthy to be admitted under the residual hearsay rule should therefore be aware of these facts and of the relative degree to which the evidence offered is prone to risks like those under discussion. Survey evidence, for example, is susceptible not only to all four classic hearsay dangers but also to an additional class of risk arising from the fact that parties usually offer surveys to support statistical inferences. These inferences can be subject to methodological error and can sometimes be manipulated through artful data collection or presentation. Proper survey methodology can, however, help ensure the reliability of the statistical inferences for which a survey is offered. Proper survey methodology can also help reduce two of the four classic hearsay dangers. In particular, the risk of insincerity can ordinarily be reduced if the interviewers and those questioned lack knowledge of the litigation and the purpose of the survey. 4 See Zippo, 216 F.Supp. at 684 (“[The] danger [of insincerity] is minimized by the circumstances of [a] public opinion poll in which scientific *234 sampling is employed, because members of the public who are asked questions about things in which they have no interest have no reason to falsify their feelings”)- Similarly, surveyors can reduce the risk of faulty narration by framing questions in a clear, precise and non-leading manner. See id. (“[0]ther survey techniques substantially insure trustworthiness in other respects. If questions are unfairly worded to suggest answers favorable to the party sponsoring the survey, the element of trustworthiness in the poll would be lacking. The same result would follow if the interviewers asked fair questions in a leading manner.”). The only risks that proper survey methodology does not tend to mitigate are those of faulty memory and perception. Still, a particular memory survey, which, for example, relates to events that were learned by direct perception and are unlikely to be forgotten, can, if properly conducted, minimize all five of the classes of risks ordinarily associated with survey evidence. See generally, e.g., Debra P., 730 F.2d at 1405; Pittsburgh Press, 579 F.2d at 751. Memory surveys can thus, in principle, have even greater circumstantial guarantees of trustworthiness than many of the traditional exceptions to the hearsay rule. Because the residual hearsay rule requires an initial trustworthiness determination before it will allow for the admission of evidence, all of these considerations, including those of methodology, will affect not only the weight but also the admissibility of surveys offered in the present circumstances, ie., when none of the other traditional hearsay exceptions apply. Contrary to Schering’s position, courts deciding whether to admit surveys in these circumstances should therefore examine their trustworthiness — and ultimately their weight, if admitted — both in terms of their methodological strengths and in terms of their proneness to faulty memory and perception. A review of the case law suggests that this totality of factors has in fact come into play when determining the use of such surveys. For example, although Schering claims that in a “litany of casés” courts have simply admitted state of mind surveys for the truth of the matters remembered or believed, only two of the many cases cited — Keith v. Volpe, 858 F.2d 467, 479-81 (9th Cir.1988) and Debra P. v. Turlington, 730 F.2d 1405 (11th Cir.1984)— actually did so over a hearsay objection, 5 *235 and both of these cases involved surveys where the risks of faulty memory and perception were relatively small. . Thus, in Keith, 858 F.2d at 467, the Ninth Circuit affirmed a district court’s decision to admit a survey that asked respondents about their race, income and housing preferences. See id. at 479-81. In determining that the survey was trustworthy, the court examined the survey’s methodology and never explicitly mentioned the risks of faulty memory or perception. Statements concerning a de-clarant’s race and income are, however, relatively unsusceptible to these two classes of risk, and this fact may have had some bearing on the outcome. 6 Similarly, in Debra P., 730 F.2d at 1405, the Eleventh Circuit affirmed a district court’s decision to admit a survey that asked teachers “whether [they] had provided instruction during 1981-82 relating to the skills tested on the SSAT-II and if so, whether that instruction had been sufficient for a student to master the skills.” Id. at 1408. In finding that the surveys were sufficiently trustworthy to be admitted, the court again examined only the factors that define a properly conducted survey. The risk that teachers would incorrectly perceive or remember whether they had taught a course in the previous year was, however, also presumably small. In Pittsburgh Press, 579 F.2d at 751, by contrast, the court excluded a memory survey after assessing whether its methodology was sound. See id. at 758-60. The Third Circuit observed: In the context of polls and surveys, the circumstantial guarantees of trustworthiness are for the most part satisfied if the poll is conducted in accordance with generally accepted survey principles, and if the results are used in a statistically correct way, since proper survey and statistical methods are intended to assure a poll’s reliability. Id. at 758 (emphasis added). The court ultimately rejected the survey, however, not only because of its methodological problems but because “[b]oth hearsay dangers — faulty memory and insincerity— loom[ed] large in the ... poll.” Id. at 759. The risk of insincerity was due to flaws inherent in the survey’s methodology, but the risk of faulty memory arose from the fact that “[t]he respondents were not being asked for a present impression; rather they were being asked for details about banquets which had taken place many years before. In fact the covering letter, recognizing that the members’ memories might be ‘dimmed or imperfect,’ nevertheless asked for a ‘best recollection.’ ” Id. (emphasis added). Thus, the surveys in Pittsburgh Press, which were rejected, polled for memories about events that were difficult to perceive and remember, while the surveys in Keith and Debra P., which were admitted, polled for memories about events that were unlikely to be mis-perceived or misremembered. In the present case, the district court did not make any findings concerning the surveys’ methodological strengths or weaknesses. Instead, it rejected them on the ground that *236 the surveys seek to recover, from the physicians’ memories, what the Pfizer or UCB representatives said to them, during a very brief personal interchange, thus interjecting a layer of uncertainty, because either of understandable inattention or distraction during that interchange or faulty memory [that is not present in state-of-mind surveys]. Schering, 1999 WL 144921, at *4. Four of the five surveys in this case were, however, performed within a day of the detailings, and the remaining one was performed within a week. Additionally, because the record shows that somnolence level was a critical factor in deciding whether to prescribe antihistamines, the physicians polled presumably would have been poised to look for and remember this type of information in a detailing. Thus, the risks of faulty memory and perception would appear to be quite marginal in this case, and the hearsay here is far more akin to that admitted in Keith and Debra than that excluded in Pittsburgh Press. Moreover, even if these surveys bear some risks of faulty memory and perception, their overall reliability cannot be compared to that of the traditional hearsay exceptions without first assessing their methodological strengths and weaknesses. For these reasons, we vacate the district court’s initial trustworthiness findings and remand for reevaluation on the basis of the surveys’ methodological strengths as well as their relative susceptibilities to the risks of faulty memory and perception. This inquiry will determine not only whether any of the five surveys should be admitted to show what Zyrtec agents were saying, but also the weight of any evidence thus admitted. We note again, however, that in assessing the reliability of these surveys, the district court should view the evidence in context. For example, all five surveys tended to corroborate one another, and one was produced at Pfizer’s request. Schering has also produced some independent evidence in support of its position, in the form of Pfizer’s' training manuals, and this evidence may further corroborate some of the surveys’ findings. Schering has, however, failed to present even one individual interviewee for cross-examination. As noted earlier, this process may have helped relieve some .of the concerns Pfizer has raised regarding the methodological validity of the surveys, and some of these concerns may, in fact, be substantial. All of these factors should be considered on remand. ii. Necessity of the Surveys The district court’s second ground for refusing to admit the five surveys under the residual hearsay rule was based on its finding that the surveys were not “more probative on the point for which [they were] offered than any other evidence which [Schering could] procure through reasonable efforts.” Fed.R.Evid. 807(B). In this case, Schering introduced its five surveys, which polled a total of approximately 1166 physicians, to establish statistical facts about what Zyrtec representatives were saying to a much larger group of almost 250,000 physicians nationwide. These facts were relevant to establish the type of irreparable injury necessary for a preliminary injunction. See, e.g., Indianapolis Colts, 34 F.3d at 416; Reuters Ltd. v. United Press Int’l, Inc., 903 F.2d 904, 907-09 (2d Cir.1990). The reliability of this statistical inference was therefore critical to Schering’s case. As the district court noted, it would, moreover, be unreasonable to hale all 1166 physicians into court to establish this kind of inference. Schering, 1999 WL 144921, at *5 (holding that this number is “without doubt too large a number to bring into Court”). Rule 807(B)’s so-called “necessity” criterion nevertheless requires courts to compare the reliability, or trustworthiness, of a survey with that of other evidence that a proponent might reasonably obtain to establish the same fact. See Zippo, 216 F.Supp. at 683 (noting that *237 inquiry “requires a comparison of the probative value of the survey with the evidence, if any, which as a practical matter could be used if the survey were excluded”). The district court’s erroneous belief that memory surveys are per se untrustworthy thus logically compelled its finding that the survey evidence in this case was unnecessary. For these reasons, we vacate this finding as an abuse of discretion and remand for reevaluation on the basis of the considerations set forth in this opinion. Perhaps because the court deemed the surveys per se unreliable, it did not rule conclusively as to whether any reasonable alternatives to the survey evidence were available in this case. The court did, however, note “without passing on” three possibilities: [1] the availability of depositions by telephone, which would minimize inconvenience to the physicians, [2] the fact that the inquiry could reasonably be limited to the physician’s recollection of what was said to him during a very brief and recent detailing, and [3] the apparent probability that a scientifically constructed random sample of a much smaller number of persons than 1166, if they could be cross-examined, would be more rehable, ie., would have greater circumstantial guarantees of trustworthiness, than hearsay. Schering, 1999 WL 144921, at *5 (citation omitted). We briefly address these possibilities here because the foregoing analysis is likely to be revisited on remand. First, although parties can indeed consent to depositions by telephone, the use of deposition testimony would most likely suffer from some of the risks of statistical error that proper survey methodology can help to avoid. In Zippo, Judge Feinberg thus noted that: [t]he alternatives of having a much smaller section of the public testify ... or using expert witnesses to testify ... are clearly not as valuable because the inferences which can be drawn from such testimony to [the statistical facts sought to be proven] are not as strong or as direct as the justifiable inferences from a scientific survey. 216 F.Supp. at 684. Second, in recommending that Schering obtain evidence limited to a “physician’s recollection of what was said to him during a very brief and recent detailing,” the district court ignored the fact that most of the surveys were conducted within a day, and one within a week, of these meetings. The district court’s criticisms would thus seem inapplicable to many, if not all, of these surveys. Finally, we are skeptical of the court’s suggestion that a smaller number of physicians, if subjected to cross-examination, might be -used to prove the same statistical points for which the surveys were offered. To make the alternative of summoning a reduced number of physicians into court reasonable, this number would have to be much smaller than 1166. It is difficult to see how such a small sample would provide more reliable evidence than the surveys, particularly when is not at all clear that the surveys were untrustworthy to begin with in ways that cross-examination could have cured. Assuming the surveys were trustworthy enough to be admitted under the residual hearsay rule, we therefore find nothing in the record to suggest that this evidence was any less necessary than in cases where surveys have been admitted. See, e.g., Grotrian, 523 F.2d at 1340; Zippo, 216 F.Supp. at 684. iii. General Interests of Justice The third ground that the district court cited for refusing to admit the five surveys under the residual hearsay rule was that “the general purposes of [the Federal Rules of Evidence] and the interests of justice [would not] best be served by admission of the [physicians’] statements] into evidence.” Fed.R.Evid. 807(C). In explaining this finding, the court said: *238 [A]dmission of [Sehering’s] surveys to prove what was said to the physicians by the Pfizer and UCB representatives would, in the Advisory Committee’s phrase, be a step toward “the virtual destruction of the hearsay rule.” Such a precedent could well lead towards a rule that the admissibility of hearsay evidence turns on the proponent’s convenience, or, perhaps, business sensibilities. Schering Corp., 1999 WL 144921, at *5 (citation and footnote omitted). As this quotation shows, this finding, like the others, rested on a per se rule against admission of memory statements. We have, however, rejected this rule as erroneous. See Section IV(i), supra. It is worth noting that we have often admitted surveys that are relevant to prove material facts under the residual hearsay rule on the basis simply of their “need ... plus adequate guarantees of trustworthiness.” Grotrian, 523 F.2d at 1341 (noting that surveys are sometimes admitted on this basis); see also 5 Jack B. Weinstein & Margaret A. Berger, Weinstein’s Evidence ¶ 901(b)(9) [03] at 901-140 (1995) (“The admissibility of survey or sampling results depends upon two factors: necessity and trustworthiness.”). In the context of survey evidence, the interests of justice and the general purposes of the rules of evidence are generally best served by the admission of surveys that meet these two criteria. We therefore vacate and remand for reconsideration of the interest-of-justice finding following reconsideration of necessity and trustworthiness. V. The Party Admission Theory: Admissibility of the Gengler Analysis and FasTape Survey Schering argues that the district court should have admitted the FasTape Survey, which Pfizer commissioned, and Pfizer’s internal analysis and summary of this survey (the “Gengler Analysis”) as party admissions under Rule 801(d)(2). We agree. Rule 801(d)(2) excludes from the definition of hearsay any statement that is offered against a party and is (A) the party’s own statement, in either an individual or a representative capacity or (B) a statement of which the party has manifested an adoption or belief in its truth, or (C) a statement by a person authorized by the party to make a statement concerning the subject, or (D) a statement by the party’s agent or servant concerning a matter within the scope of the agency or employment, made during the existence of the relationship.... Fed.R.Evid. 801(d)(2). “Admissions by a party opponent are excluded from the category of hearsay on the theory that their admissibility in evidence is the result of the adversary system, rather than satisfaction of the conditions of the hearsay rule.” Fed.R.Evid. 801(d)(2) advisory committee’s note (citing Edmund M. Morgan, Basic Problems of Evidence 265 (1962); John S. Strathorn, A Reconsideration of the Hearsay Rule and Admissions, 85 U. Pa. L.Rev. 484, 564 (1937); 4 John Henry Wigmore, Evidence in Trials at Common Law § 1048 (Chadbourn rev.1972)). The Advisory Committee thus recommends “generous treatment of this avenue to admissibility.” Id. With regard to the Gengler Analysis, Schering correctly points out that this document was drafted by a Pfizer employee, Lakshmi Gengler, when she was a member of Pfizer’s market research and analytic team. Gengler’s job during this period was to work with [this] team to understand which market research the team would like to conduct in support of its business, and then identify] vendors who [could] conduct that work and work with the team to develop the questionnaires to be used in that work, and then [have] the vendor conduct the work and then help[ ] with the analysis of the work. *239 (J.A. at 59.) Gengler helped design and commission the FasTape Survey in this capacity. Because the statements in the Gengler Analysis were made “by [a Pfizer] agent or servant concerning a matter within the scope of the ... employment, made during the existence of the relationship,” they qualify as party admissions under Rule 801(d)(2)(D). The Gengler Analysis, however, partly summarized the FasTape Survey, which itself contained hearsay in the form of physicians’ out-of-court descriptions of de-tailings. To be admitted, this document must thus survive a double hearsay analysis. See, e.g., Pittman v. Grayson, 149 F.3d 111, 124 (2d Cir.1998). In Pittman, for example, we held that a party admission was “properly excludable on the ground that it contained hearsay.” Id. We noted that the party who made the admission “stated she was repeating a story she had heard from someone else.” Id. Schering, however, points out that in Pekelis v. Transcontinental & Western Air, 187 F.2d 122 (2d Cir.1951), by contrast, we admitted a similar party admission in circumstances where [t]he reports[, which were party admissions,] did not merely repeat what was told the board by witnesses, but drew inferences as to what in fact happened and those inferences were contrary to the position taken by the defendant during the trial. Id. at 129 (A.Hand, J.). Schering argues that Pefceiisstands for the proposition that “well before the enactment of the Federal Rules [of Evidence], this Court recognized that the fact that an admission contains hearsay is ‘no valid objection....”’ (Pl.’s Br. at 59.) Pekelis and Pittman are, however, better harmonized on the theory that a party admission containing hearsay is admissible where, as in Pekelis, the admission draws inferences from the underlying hearsay and thus “manifest[s] an adoption or belief in its truth.” Fed.R.Evid. 801(d)(2)(B) (defining such statements as party admissions). A party admission may, however, be inadmissible when it merely repeats hearsay and thus fails to concede its underlying trustworthiness. In the present case, the district court viewed the Gengler analysis as “no more than a convenient summary of the research, not representing a conclusion by Pfizer.” Schering, 1999 WL 144921, at *7. The Gengler Analysis stated, however, that [flour of the top 5 messages communicated to physicians for Zyrtec are completely appropriate and relevant.... The fifth and less appropriate message being communicated by Zyrtec representatives concerns their claim of Zyrtec causing little or no sedation to allergic rhinitis patients. (Emphasis added). To be able to state generally that a message is “being communicated by Zyrtec representatives” on the basis of the small number of prior statements recorded in the FasTape Survey, Gengler had to believe that this data could support an inductive inference that Zyrtec’s message was being communicated more broadly. This belief in turn required reliance on the trustworthiness of the physicians’ statements and on the survey methodology itself. By making this inference, Gengler thus manifested a belief in the trustworthiness of both and conceded the survey’s reliability. In fact, Gengler specifically approved the FasTape Survey’s methodology before the research was performed. Pekelis controls in these circumstances, and both the Gengler Analysis and the FasTape Survey are admissible over a hearsay objection. CONCLUSION For the foregoing reasons, we vacate the district court’s rulings excluding the five surveys in this case. We hold that the Scott Levin Survey is admissible over a hearsay objection under Rule 803(3) for the limited purpose of showing what Zyrtec representatives were implying in the detailings. We hold that the FasTape *240 Survey is admissible both to show these implications under Rule 803(3), and as a party admission under Rule 801(d)(2) to show what the representatives were actually saying. We remand to the district court for further proceedings to determine whether any of the surveys are sufficiently trustworthy and necessary to be admitted under Rule 807 to show literal falsehoods in violation of the Settlement Agreement and Section 43(a)(2) of the Lanham Act. We express no opinion, however, as to whether there are any other grounds for excluding the above evidence, and we leave it to the district court to determine what weight to give all of the evidence admitted on remand. Footnotes [1] This panel appears to have been chosen to be representative of all such physicians nationwide. [2] Significantly, the Settlement Agreement was entered into after Pfizer and UCB received a notice from the FDA, stating that Pfizer’s sales representatives may be promoting Zyrtec as a "low sedating” or “nonsedating” antihistamine. Because [the FDA] considers either of these claims presents a safety concern in light of Zyrtec's approved product labeling ... [the FDA] recommend[s] that Pfizer review the promotional activity of its sales representatives and have its sales representatives cease any promotional activity that might suggest or imply that Zyrtec is a “low sedating” or "nonsedating” antihistamine. (J.A. 467 (emphasis added).) [3] Not all courts examining expressions of mental states have deemed such expressions to be hearsay, as we are suggesting here. In Zippo, the court explained that [t]he cases holding that [state-of-mind] surveys are not hearsay do so on the basis that the surveys are not offered to prove the truth of what respondents said and, therefore, do not fall within the classic definition of hearsay. 216 F.Supp. at 682 (footnotes omitted). The court also noted, however, that: [t]his approach has been criticized because, it is said, the answers to questions in a survey designed to prove the existence of a specific idea in the public mind are offered to prove the truth of the matter contained in these answers. Under this argument, when a respondent is asked to identify the brand of an unmarked lighter, the answer of each respondent who thinks the lighter is a Zippo is regarded as if he said, "I believe that this unmarked lighter is a Zippo.” Since the matter to be proved in a secondary meaning case is respondent's belief that the lighter shown him is a Zippo lighter, a respondent's answer is hearsay in the classic sense. Others have criticized the non-hearsay characterization, regardless of whether surveys are offered to prove the truth of what respondents said, because the answers in a survey depend for their probative value on the sincerity of respondents. Id. at 682-83 (footnotes omitted). We believe that state-of-mind expressions are hearsay even if they are offered only to establish the existence of the relevant states of mind. Indeed, Rule 803(3), which creates an exception to the hearsay rule for such statements rather than excluding the statements from the definition of hearsay, makes sense only in this light. [4] In cases where interviewees are likely to lie for personal reasons, anonymous surveying can help minimize this risk. [5] Three of the cases cited as admitting memory surveys are inapposite because they involve review of either administrative or congressional decisions, both of which can be based on evidence that would be inadmissible in a court of law. See Turner Broadcasting System, Inc. v. FCC, 520 U.S. 180, 181, 195, 202-03, 117 S.Ct. 1174, 137 L.Ed.2d 369 (1997) (holding that "Congress has drawn reasonable inferences based on substantial evidence,” when it had before it evidence including some memory surveys); Motor Vehicle Mfrs. Ass’n v. State Farm Mutual Auto. Ins. Co., 463 U.S. 29, 52, 103 S.Ct. 2856, 77 L.Ed.2d 443 (1983) (allowing admission of memory surveys to support the National Highway Traffic Safety Administration's rescission of crash protection requirements, though finding this evidence insufficient); Bristol-Myers Co. v. FTC, 185 F.2d 58, 60 (4th Cir.1950) (holding that FTC action was supported by substantial evidence, which included a memory survey). Three other cases relied on by Schering are irrelevant for technical reasons. See Bank of Utah v. Commercial Sec. Bank, 369 F.2d 19, 27-28 (10th Cir.1966) (refusing to address whether the surveys were admissible hearsay because the surveys were clearly inadmissible on technical grounds); Indian Head, Inc. v. Allied Tube & Conduit Corp., 817 F.2d 938, 939 & n. 1 (2d Cir.1987) (referencing a survey in a footnote, but only to support a proposition to which both parties agreed); Sony Corp. v. Universal City Studios, Inc., 464 U.S. 417, 423, 104 S.Ct. 774, 78 L.Ed.2d 574 (1984) (admitting memory surveys, but only when surveys submitted by both parties showed the same facts, when neither party raised hearsay objections, and when the only issue was whether the facts established by these surveys were sufficient to establish liability). Two other cases cited as admitting memory surveys are better described as admitting only state-of-mind surveys. See Harolds Stores, 82 F.3d at 1545-46 (asking — apart from questions designed to establish universe of survey — only about dispositions to shop at a certain store, which are state of mind state *235 ments); Randy’s Studebaker Sales, Inc. v. Nissan Motor Corp., 533 F.2d 510, 520 (10th Cir.1976) ("We are of the opinion that the questionnaires were properly admitted to reflect the then existing state of mind of the customers as to the quality of Randy’s services generally [i.e., to establish customer satisfaction].”). Finally, the remaining cases cited by Scher-ing involved surveys collecting statements of both remembered events and then-existing states of mind where the memory statements were used only to establish the universe of the survey. (See Pl.'s Br. at 42-43.) Because these surveys were used to establish the truth of the state of mind propositions, and not the universe of the study, the surveys would have been independently admissible under 803(3), and the problems with the universe statements properly went to the weight of the surveys. [6] Statements concerning one's housing preferences would, in turn, presumably be independently admissible under Rule 803(3), as presently-existing states of mind.