Full opinion text
OPINION AND ORDER J. PAUL OETKEN, District Judge. Plaintiff Reed Construction Data, Inc. (“Reed”) brought this action against Defendants McGraw-Hill Companies, Inc., (“McGraw-Hill”), five unidentified natural persons, and five unidentified entities, claiming violations of the Lanham Act, 15 U.S.C. §§ 1114(l)(a), 1125(a)(1)(A), the Sherman Antitrust Act, 15 U.S.C. § 2, and various state law torts. Reed alleges that McGraw-Hill surreptitiously accessed Reed’s database service and used that access to generate false or misleading product comparisons that it distributed to prospective Reed customers. McGraw-Hill has moved for summary judgment and has moved to exclude the testimony of Reed’s expert witness, Dr. Frederick Warren-Boulton. The Court conducted a two-day Daubert hearing on the admissibility of that testimony on July 30 and 31, 2014. For the reasons that follow, McGraw-Hill’s motion to exclude is granted, and its motion for summary judgment is granted in part and denied in part. I. Background The parties are in the business of providing construction product information (“CPI”). CPI began at the turn of the century with written newsletters that provided information on construction projects so that subscribers—ordinarily those in the building trades—could bid for jobs. See F.W. Dodge Co. v. Construction Info. Co., 183 Mass. 62, 66 N.E. 204 (1903). Today, CPI services are nationwide searchable databases that can filter projects based on the user’s preferences. For example, a user can search for library projects in Topeka, Kansas, worth more than three million dollars, that need plumbing in the next two months. The CPI service will collect all the projects meeting those specifications and provide plans, bidding information, and contact information for the planner, architect, or general contractor on the job. Reed and McGraw-Hill each sell CPI subscriptions at the national, state, and local levels. McGraw-Hill’s service is called the “Dodge Network.” Reed’s is called “Connect.” At the national level, Reed and McGraw-Hill are the only two services in the market. CPI customers generally prefer a service that lists more projects to one that lists fewer projects, all else being equal. The parties, knowing this, compete vigorously over who has the most projects in its database. They also endeavor to protect the information in their databases from unauthorized use. The user agreements that Reed’s and McGraw-Hill’s customers must sign limit the permissible uses of the information in the database, which do not include creating comparisons with competing CPI providers. Around 2004, McGraw-Hill began to access Reed Connect for two purposes. First, McGraw-Hill wanted to create comparisons that it believed would be favorable to its Dodge Network. To do this, it needed to know how many projects were listed on Reed Connect. Second, McGraw-Hill wanted to be aware of changes in the marketplace and ensure that Reed was not listing significant projects that it had missed. To do this, McGraw-Hill needed to know what projects Reed was listing. McGraw-Hill endeavored to conceal that it was subscribing to Reed Connect. McGraw-Hill paid consultants—referred to internally as “spies”—to subscribe to Reed Connect. The consultants would create fake entities to subscribe to Reed Connect and would sometimes falsely claim that those entities were associated with actual builders and contractors. McGraw-Hill paid these consultants in cash and personal checks drawn on the McGraw-Hill credit union and listed the expenses on its books as “Stationery and Supplies,” or “Magazines and Books.” Ultimately, McGraw-Hill spent $3.45 million on consultants’ accessing Reed Connect. Once it had access to Reed Connect, McGraw-Hill hired GFK Roper Public Affairs & Media, Inc., (“Roper”) to generate product comparisons. Roper advertised itself as an independent entity evaluating the two services. But, according to Reed, Roper did little more than send someone to sit in a room and watch a McGraw-Hill employee run searches on the two services. Roper made no effort to ensure that the two searches were fairly comparable. For example, McGraw-Hill actually used one of its other (presumably superior) products in the tests but said that it had used the Dodge Network. Further, the searches were selected so as to emphasize McGraw-Hill’s strengths and minimize those of Reed. Because Reed had superior listings for projects worth under $1 million, McGraw-Hill limited the comparisons in the Roper Reports to projects worth more than $1 million. Similarly, McGraw-Hill ran the searches so as to capture projects that needed to be completed expeditiously—these projects are called “ASAPs”—in its database but not to capture them in Reed’s database. These comparisons resulted in a report in which McGraw-Hill boasted “71% more planning projects, 78% more bidding projects, and 71 % more digitized plans and specifications.” (Plaintiffs Exhibit 484 at 33.) Ultimately, Reed had its own expert analyze the data, and came to the conclusion that the Roper reports were biased in McGraw-Hill’s favor. While McGraw-Hill was distributing the Roper reports, it was also conducting ad hoc comparisons of the two services in response to questions from customers. According to Reed, McGraw-Hill issued 1,235 unique ad hoc comparisons based on its unauthorized access to Reed Connect. When customers wanted to compare the services, McGraw-Hill frequently advised them to search for a particular project in both services, knowing all the while that the suggested project would be found only in the Dodge Network. McGraw-Hill also issued a number of state and local comparisons of the two products that were generally similar to the Roper reports in both content and methodology. At the same time, McGraw-Hill touted a five-to-one advantage in “exclusive” projects—those that McGraw-Hill covered but Reed did not— in its communications with customers, particularly large customers. In reality, Reed contends, the true ratio was closer to 2.6-to-one. On at least a few occasions, McGraw-Hill used its access to Reed Connect to find new projects. Reed describes this as “stealing.” (Dkt. No. 156, Plaintiffs Memorandum of Law at 35 [hereinafter “Plaintiffs Memorandum”].) McGraw-Hill describes it as “isolated potential violations of McGraw-Hill’s rules in which McGraw-Hill may have used Reed Connect to obtain a source of project leads.” (Dkt. No. 150, Defendant’s Memorandum of Law, at 15 n. 6 [hereinafter “Defendant’s Memorandum”].) McGraw Hill claims that it had “strict rules” (id.) in place to regulate the use of its illicitly obtained Reed Connect access. (These rules were, fittingly, called the “Roper Rules.”) The parties agree that McGraw-Hill broke these rules at least a few times and used its access to Reed Connect for purposes other than generating comparisons. Reed brought suit against McGraw-Hill in 2009, pressing its current claims as well as claims under the Racketeering-Influenced and Corrupt Organizations Act (RICO). Judge Sweet dismissed the RICO claims, but allowed the remaining claims to proceed. During the motion to dismiss stage, Reed alleged that no fewer than 231 customers reported noticing the Roper Reports and were influenced by their contents. Discovery has not borne out that claim. Reed can now point to one customer declaration showing that the Roper Reports influenced purchasing decisions. In addition to arguing that it was injured by losing customers, Reed now claims that it was injured because it was forced to price its services lower than it otherwise would have absent McGraw-Hill’s misconduct. In support of this damages theory, Reed offers the testimony of Dr. Frederick Warren-Boulton. McGraw-Hill has moved to strike that testimony under Federal Rule of Evidence 702. II. Motion to Exclude Dr. Warren-Boulton’s Testimony Reed has retained Dr. Warren-Boulton to answer four questions related to this case. First, is there a distinct national market for CPI sufficient to trigger § 2 of the Sherman Act? 15 U.S.C. § 2. Second, did McGraw-Hill exercise power in that market? Third, did McGraw-Hill’s misconduct allow it to keep its market power? And, finally, did McGraw-Hill’s misconduct damage Reed? The parties refer to Dr. Warren-Boulton’s answers to the first three questions as his “liability opinion” and his answer to the final question as his “damages opinion.” A. Regression Analysis To support both his liability and damages opinions, Dr. Warren-Boulton conducted statistical regression analyses of Reed’s and McGraw-Hill’s pricing and service data. A brief overview of regression analysis may be helpful. The basic regression method is simple: isolate the effect of one variable (the “independent variable”) on another variable (the “dependent variable”) by holding all other potentially relevant variables (the “control variables”) constant. By controlling for other factors that might influence the dependent variable, one “regresses” the influence of the independent variable on the dependent variable. The number associated with that influence is called a “coefficient.” Imagine, for example, that one wanted to isolate the effect of location on the price of an apartment. One would start by comparing the prices of apartments (the dependent variable) of the same size, with the same number of bathrooms, amenities, etc. (the control variables), across different locations (the independent variable). Regression analysis formalizes that method by solving an equation of the dependent variables with the independent and control variables for the linear function that best fits the data. To visualize this, the regression model finds the line that best fits a plot of the data points. The spaces between the line and the actual data points are called “residuals.” Once the line is found, the analyst must test the validity of the results. To do this, analysts run a series of mathematically complicated tests to answer two uncomplicated questions: (1) How well do the data fit the model? And (2) are the residuals (the spaces between the data points and the regression line) significantly correlated with any of the control variables or the independent variable? To simplify: if the answer to (1) is “not well,” the analyst has a problem of statistical significance; if the answer to (2) is yes, she has probably omitted an important variable from her model. The fundamental goal of regression analysis is to convert an observation of correlation (e.g., apartments in Manhattan cost more than those in Queens) into a statement of causation (apartments in Manhattan cost more than those in Queens because they are in Manhattan, not because they are larger or more luxuriously appointed). Models called “residual models” attempt to do this by controlling for every observable variable that might have an effect on the dependent variable and seeing if the residuals are significantly correlated with an explanatory variable. Dr. Warren-Boulton’s is a residual model. B. Dr. Warren Boulton’s Method Dr. Warren-Boulton was charged with determining whether McGraw Hill’s “misconduct” damaged Reed. Reed has already conceded, based on the evidence following extensive discovery, that only a handful of individual customers relied on McGraw-Hill’s allegedly false product comparisons. However, Warren-Boulton found what he calls a “price effect.” He hypothesized that customers paid more for McGraw-Hill’s service than they otherwise would have because of McGraw-Hill’s misconduct. He further hypothesized that the amount of McGraw-Hill’s gain was the amount of Reed’s loss—that where McGraw-Hill’s prices were inflated, Reed’s were deflated. To isolate the price effects of McGraw-Hill’s misconduct, Warren-Boulton proposed a “benchmark” model, which is a form of residual model where unobservable data (misconduct, in this case) is extrapolated by comparing observable data to a known benchmark statistic. Warren-Boulton’s benchmark model compares the parties’ prices for national services during the relevant period with the parties’ prices for local services during the relevant period. He calls the ratio of national to local pricing (for each party) the “price index ratio.” To make the benchmark model work, Warren-Boulton assumes that national pricing is affected by McGraw-Hill’s misconduct significantly more than local pricing is and that the effects of McGraw-Hill’s misconduct will grow weaker over time (because the misconduct ceased in approximately 2008). The data are indexed to a fixed time in 2013 by which Warren-Boulton assumes that the effects of McGraw-Hill’s misconduct will have fully dissipated. Under Warren-Boulton’s model, then, if the difference between each party’s price index declines over the relevant period, and that decline cannot be attributed to any other observable factor, then there is proof that McGraw-Hill’s malfeasance worked a price effect. The parties have likened this expected effect to a funnel: at the start of the misconduct, the gap between the price indices is wide; by the end, it has narrowed to nothing. To test his hypothesis, Warren-Boulton constructed two regression models that control for as many relevant variables as he could think of and measure (what kind of service was being provided, the coverage area, the relevant markets, etc.) and seek to isolate the effect of the length of time since the relevant period began (the independent variable) on the price index ratios (the dependent variable). Time is measured by a variable indicating each quarter since the relevant period began. If the coefficients associated with the quarterly variables are significant and negative, Dr. Warren-Boulton’s hypothesis would be confirmed. Controlling for everything else, Dr. Warren-Boulton could claim strong statistical proof for the “funnel” theory of damages. C. Regression Analysis under Rule 702 The admission of expert evidence is governed by Federal Rule of Evidence 702, which codified the Supreme Court’s holding in Daubert v. Merrell Dow Pharmaceuticals, 509 U.S. 579, 113 S.Ct. 2786, 125 L.Ed.2d 469 (1993). The rule charges district courts with determining whether (1) “scientific, technical, or other specialized knowledge will assist the trier of fact,” (2) the expert is qualified “by knowledge, skill, experience, training, or education” to testify on that subject, (3) the expert’s proffered testimony is grounded on “sufficient facts or data,” (4) that testimony is the product of “reliable principles and methods,” and (5) the expert “applies the principles and methods reliably to the facts of the case.” See also Bricklayers & Trowel Trades Int’l Pension Fund v. Credit Suisse First Boston, 853 F.Supp.2d 181, 186 (D.Mass.2012), aff'd sub nom. Bricklayers & Trowel Trades Int’l Pension Fund v. Credit Suisse Sec. (USA) LLC, 752 F.3d 82 (1st Cir.2014). Courts consider several factors in determining whether testimony is sufficiently scientific to be admissible, including “whether the theory or technique can be and has been tested; (2) whether the technique has been subject to peer review and publication; (3) the technique’s known or potential rate of error; and (4) the level of the theory or technique’s acceptance within the relevant discipline.” Bricklayers, 752 F.3d at 91 (internal citations omitted). The burden is on the proponent of the expert’s testimony to prove that it is admissible. Moore v. Ashland Chem. Inc., 151 F.3d 269, 276 (5th Cir.1998) (“The proponent need not prove to the judge that the expert’s testimony is correct, but she must prove by a preponderance of the evidence that the testimony is reliable.”). But “the test of reliability is ‘flexible,’ and Daubert’s list of specific factors neither necessarily nor exclusively applies to all experts or in every case.” Kumho Tire Co., Ltd. v. Carmichael, 526 U.S. 137, 141, 119 S.Ct. 1167, 143 L.Ed.2d 238 (1999). The Court’s task under Daubert is to determine whether the proffered methodology constitutes “good science” and, therefore, whether it produces “scientific ... knowledge.” Daubert, 509 U.S. at 593, 113 S.Ct. 2786; Fed.R.Evid. 702. Scientific knowledge is different from ordinary knowledge. See KaRL Poppek, the LogiC of SCIENTIFIC DISCOVERY (1934). And knowledge is different from other classes of belief. Cf. Bertrand Russell, On Denoting, 14 Mind 479, 479-93 (1905). The standard, then, is formally higher than the standard that prevailed before Daubert, which required only that the offered testimony be the result of a method that was generally accepted in the relevant analytical community. Frye v. United States, 293 F. 1013, 1014 (D.C.Cir.1923). But Daubert also opened new avenues for scientific evidence to be admitted. Some “good science” has not yet been accepted by the relevant community. The mere fact that experts would reject the proffered testimony—indeed, the mere fact that most experts would reject the proffered testimony—is alone insufficient to render that testimony inadmissible. See, e.g., 3 Mueller & KirxpatriCK, Federal Evidenoe § 33:6-8 (3d ed.2007). The Court must independently serve as the “gatekeeper” for the finder of fact and is left with the “heady task” of “occasionally ... rejecting] expert testimony because it was not derived by the scientific method.” Daubert v. Merrell Dow Pharm., Inc., 43 F.3d 1311, 1316 (9th Cir.1995) (Kozinski, J.) [hereinafter “Daubert II ”], on remand from Daubert, 509 U.S. 579, 113 S.Ct. 2786. Courts, though, must not determine the credibility of the expert’s proffered testimony or compare two experts for the purpose of determining which of them is correct. Rather, Daubert and Rule 702 instruct courts to exclude only testimony that is unscientific or unlikely to assist the trier of fact in the determination of a relevant issue. E.g., Ruiz-Troche v. Pepsi Cola of Puerto Rico Bottling Co., 161 F.3d 77, 81 (1st Cir.1998). Strictly speaking, one need not have certainty to have knowledge. Daubert II, 43 F.3d at 1316; see also Don Herzog, Cute Prickly Critter with Presbyopia, 110 Mich. L.Rev. 953, 957 (2012) (reviewing Ronald DworKIN, JuSTICE FOR HEDGEHOGS (2011)). The possibility of mistake need only be tolerable, not remote. Within that tolerable range, disputes over competing expert opinions are to be resolved by the trier of fact. Outside it, courts exclude the testimony. In the context of regression analysis testimony, some (slightly) more specific standards have emerged in the case law. Questions about the admissibility of regression analyses often arise in two kinds of eases: (1) securities fraud actions in which the plaintiff needs to prove loss-causation using stock-price data, and (2) employment discrimination cases in which the plaintiff needs to prove disparate impact or disparate treatment using employment data. E.g., Bazemore v. Friday, 478 U.S. 385, 400, 106 S.Ct. 3000, 92 L.Ed.2d 315 (1986) (employment discrimination); Bickerstaff v. Vassar College, 196 F.3d 435, 440 (2d Cir.1999) (discrimination); Roger v. Reno, 98 F.3d 631, 637 (D.C.Cir.1996) (discrimination); Sobel v. Yeshiva Univ., 839 F.2d 18, 20 (2d Cir.1988) (discrimination); Rossini v. Ogilvy & Mather, Inc., 798 F.2d 590, 593 (2d Cir.1986) (discrimination); Morgan v. Harris Trust & Savings Bank, 867 F.2d 1023, 1028 (7th Cir.1989) (discrimination); Allen v. Seidman, 881 F.2d 375, 378 (7th Cir.1989) (discrimination); Bricklayers, 752 F.3d at 91 (securities); In re Xerox Corp. Sec. Litig., 746 F.Supp.2d 402, 411 (D.Conn.2010) (securities); Gordon Partners v. Blumenthal, 02-CV-7377, 2007 WL 431864 (S.D.N.Y. Feb. 9, 2007), rep. and rec. adopted, 02-CV-7377, 2007 WL 1438753 (S.D.N.Y. May 16, 2007), aff'd, 293 Fed.Appx. 815 (2d Cir.2008) (securities); see also S.E.C. v. Razmilovic, 738 F.3d 14, 34 (2d Cir.2013), as amended (Nov. 26, 2013), cert. denied, — U.S.-, 134 S.Ct. 1564, 188 L.Ed.2d 561 (2014) (evaluating credibility of regression experts at bench trial on a securities fraud action). The securities cases provide guidance on (1) permissible methods to select the relevant time periods in which to observe the data and (2) the general limits of subjectivity in regression analysis. E.g., Bricklayers, 752 F.3d at 87-89. The employment discrimination cases provide guidance on which control variables are necessary for a regression analysis to be admissible. E.g., Bickerstaff, 196 F.3d at 450. First, to be admissible, a regression analysis must examine an appropriate selection of data. When constructing a benchmark statistic, the regression analyst may not “cherry-pick” the time-frame or data points so as to make her ultimate conclusion stronger. Bricklayers, 752 F.3d at 89 (internal citations omitted). Rather, some passably scientific analysis must undergird the selection of the frame of reference. Ordinarily, in the securities cases at least, the observation dates are selected by looking to the underlying facts of the case: when did the relevant events occur and cease to occur? Id. The Bricklayers case is instructive. There, the plaintiffs’ expert was retained to determine whether alleged misconduct had artificially inflated the price of the defendant’s securities. To do this, the expert needed to test the volatility of the defendants’ stock price on days when allegedly false or corrective disclosures were made. These days are called “event days.” Id. But instead of choosing dates with reference to the facts of the case itself, the analyst cherry-picked the most volatile dates in order to make the defendants’ disclosures seem more important than they were. Id. at 92. The First Circuit concluded that this was impermissible. Second, to be admissible, a regression analysis must be the product of a consistently followed methodology. Some believe that statistics is more an art than a science. Cf. Mark Twain, Chapters from My Autobiography, 186 N. AMERICAN REV. 161, 166 (1907) (expressing the view that there are “three kinds of lies: lies, damned lies, and statistics”). But for the purposes of Daubert, the practice of the art must yield to predictable and justifiable methodology. Again, Bricklayers is instructive. There, the regression analyst needed to create a benchmark model of defendants’ stock prices against which to measure the volatility of the prices on the event days. But when he constructed that background model, he excluded any day on which any news came out about the defendants’ business that could have had an effect on the stock price—as opposed to days on which allegedly corrective news came out. The district court found this to be impermissible cherry-picking as well: there was no valid reason to exclude days on which news that was not the subject of the suit was revealed. Id. But the appellate court disagreed. Id. at 93-94. It held that because the plaintiffs had presented learned scholarship supporting the proposition that the price of stock can be unusually volatile on days when material news is released, it was fair for plaintiffs’ expert to exclude those days. Whether his methodology unfairly singled out the days on which allegedly false or corrective disclosures were made was a matter for the jury to decide. Id. Finally, to be admissible, a regression analysis must control for the “major factors” that might influence the dependent variable. Bazemore, 478 U.S. at 400, 106- S.Ct. 3000. But, “[n]ormally, failure to include variables will affect the analysis’ probativeness, not its admissibility.” Id. Thus, a regression that excludes “major” variables is inadmissible while a regression that excludes other potentially important ones may not be. Bickerstaff is instructive on this point. There, the plaintiff sued the defendant—a college at which she was a professor—for failing to promote her because of her race and gender. Bickerstaff, 196 F.3d at 440. In an effort to prove that the defendant’s nondiscriminatory justification was false, plaintiff offered a regression analysis that purported to show a connection between salaries and race among the college’s faculty. The regression, though, failed to control for variations in teaching evaluations and duration of service. Id. at 449. Because these were two of the most important features upon which the college made salary decisions, their omission was fatal to the analysis. Id. Contrast this result with Bazemore. There, a regression analysis of salary data in a discrimination case was admissible even though it failed to control for county-to-county differences in salary amounts and yearly salary increases. Bazemore, 478 U.S. at 400, 106 S.Ct. 3000. Whether those omissions impermissibly biased the results was a matter for the jury to decide. D. Problems with Dr. Warren Boul-ton’s Method This Court held a Daubert hearing at which Dr. Warren-Boulton testified about his regression analysis and the opinions he derived from it. In rebuttal, McGraw-Hill called Dr. Sumanth Addanki to examine Dr. Warren-Boulton’s work and opine on its potential flaws. Neither party challenges the qualifications of the other’s expert. Reed has not challenged Dr. Addanki’s testimony under Rule 702. The Court turns to its assessment of whether Dr. Warren-Boulton’s regression analysis is admissible under Rule 702. 1. Model Design McGraw-Hill objects to Dr. Warren-Boulton’s model design in two respects. Both objections concern whether—assuming Dr. Warren-Boulton’s statistical methodology is correct—the results stand for what he claims they do. First, McGraw-Hill objects to the use of local pricing data as a baseline statistic. Second, McGraw-Hill objects to Dr. Warren-Boulton’s assumption that the market for CPI might manifest a price effect without any noticeable quantity effect. Dr. Warren-Boulton’s model is premised on the theory that McGraw-Hill’s misconduct would create a gap in the price indices that begins at the start of the misconduct and gradually narrows as the misconduct recedes into the past. This, in turn, is based on the assumption that McGraw-Hill’s misconduct worked its ill effects almost exclusively in the market for national CPI. McGraw-Hill argues that this assumption is inconsistent with the evidence in the case and with Dr. Warren-Boulton’s other findings. Dr. Warren-Boulton’s liability opinion concludes that the national market for CPI is a distinct one and is, therefore, subject to myriad different market forces, any one of which could be the cause of the narrowing gap that Dr. Warren-Boulton attributes to McGraw-Hill’s misconduct. Indeed, during his testimony he acknowledged that “Reed presumably has been becoming a more effective competitor. And so you’ve got increasing competition in the national market. But not necessarily in the local market because the local market has always been competitive.” (Dkt. No. 213, Daubert Hearing Transcript at 231 [hereinafter “Hearing Transcript.”].) If this is true—and the evidence suggests that it is—McGraw-Hill would be forced to .cut its national prices but not its local prices. This would adequately explain the narrowing gap without the presence of any misconduct. In his declaration in support of Reed’s position, Dr. Warren-Boulton responds to the criticism that local pricing is an inappropriate baseline statistic by arguing that if misconduct had an effect on both national and local prices, his model would underestimate damages because the baseline against which they are measured would be declining. The problem with this argument is that it is not responsive to McGraw-Hill’s concerns. McGraw-Hill is principally concerned that another unobserved variable—increased competition— can explain the narrowing gap. Dr. Warren-Boulton has not answered this concern—a significant one that, coupled with other flaws in his methodology discussed below, renders the model inadmissible under Daubert. Second, McGraw-Hill takes issue with a consequence of Dr. Warren-Boulton’s model. This argument takes the form of a reductio ad absurdum: if Dr. Warren-Boulton’s conclusion is true, it leads to another conclusion that must be false and, therefore, Dr. Warren-Boulton’s conclusion must be false. E.g., Leigh S. Cauman, First-Order Logic: an Introduction 36 (1998). Specifically, Dr. Warren-Boulton finds a price effect without any corresponding quantity effect: that is, he finds that the misconduct differentially affected the prices that customers were willing to pay for each of the two competitors’ services, but had no effect on how much customers chose one over the other. This goes against standard microeconomic theory, which predicts that in almost all markets, an increase in the price of a good leads to a decrease in the quantity of that good the market demands. E.g., Roger A. Arnold, Microeconomics 139 (2010). The types of goods for which this prediction does not hold true are perfectly inelastic goods, Giffen goods, and Veblen goods. Perfectly inelastic goods are defined by the fact that people will buy the same quantity regardless of their price. (Oxygen would be one, if it were for sale.) Id. Giffen goods are low-quality goods for which an upward change in price produces an upward change in quantity demanded because consumers can no longer afford superior goods. (If the price of bread goes up, those with very little money might buy more of it because they can no longer afford meat.) Alfred Marshall, Principles of Economics (1895). And Veblen goods are those characterized by conspicuous consumption. (People buy Maserati cars and Rolex watches because they are expensive.) Thorstein Veblen, the Theory of the Leisure Class: an Economic Study of Institutions (1912). Reed has not argued that CPI fits into any of these categories. So, absent a contrary explanation, the reductio holds and suggests that Dr. Warren-Boulton’s conclusion is flawed. Dr. Warren-Boulton responded to this objection by noting that the CPI market is characterized by negotiated prices. That is, there is no price set in advance by the seller; each CPI subscription is negotiated individually. This, Dr. Warren-Boulton testified, means that one could reasonably expect a price effect without a quantity effect because the price each consumer is willing to pay is a function of the price of the competing product and the relative value of the competing product and the negotiated product. Thus, Warren-Boul-ton asserts, as the perceived values of each product change, the prices will precisely change in lockstep, thereby eliminating any corresponding quantity effect. But this response fails to consider that the companies providing the services are not infinitely flexible in their price negotiations. The price that a consumer is willing to pay for Reed may well be less than the price Reed is willing to sell its service for. If that is the case, the consumer will switch to McGraw-Hill if (and only if) McGraw-Hill is willing to sell its CPI subscription at a price less than what the consumer is willing to pay. As the relative value of McGraw-Hill’s service increases and Reed’s declines, Warren-Boulton’s model predicts that the relative prices will follow the relative values, increasing the number of consumers likely to switch services. In order to accommodate the lack of a quantity effect, then, Warren-Boulton must be assuming that the range of acceptable prices for the vast majority of consumers is exceeded by the range of acceptable prices for the service providers. But there is no evidence to support that unstated assumption and no logical reason to suspect it would be true. Dr. Warren-Boulton has offered no reason to believe that the market for CPI services manifests these unusual economic characteristics—a conclusion that flows inexorably from his model. Thus, this problem remains persuasive and contributes to a finding that Warren-Boulton’s analysis is inadmissible under Daubert. 2. Omitted Variable Bias Omitted-variable problems—as the name suggests—arise when important control variables are left out of the model. Imagine trying .to calculate the effect of location on the price of an apartment without considering the size of the apartments in the sample. One might end up with what looks like a correlation between location and price, but the result would be meaningless because the entire effect could just as easily be explained by the fact that larger apartments are concentrated in certain locations. McGraw-Hill argues that Warren-Boulton’s first model (the one that omitted construction volume data) suffers from this flaw because construction volume data—which is a measurement of the overall amount of construction spending in the nationwide economy— could explain the declining gap between national and local prices. Regression analyses are admissible even where they omit important variables so long as they account for the “major variables” affecting a given analysis. Bazemore v. Friday, 478 U.S. 385, 400, 106 S.Ct. 3000, 92 L.Ed.2d 315 (1986). But it is the “proponent who must establish that the major factors have been accounted for in a regression analysis.” Freeland v. AT & T Corp., 238 F.R.D. 130, 145 (S.D.N.Y.2006). To rebut the contention that construction volume is an essential variable, Warren-Boulton claims that it could theoretically have effects in two opposite directions. On the one hand, demand for CPI services might go up when construction goes down because companies are desperate for business. On the other hand, demand might go down because construction companies have less cash on hand and fewer projects to hunt for. McGraw-Hill offers two objections to this reasoning. First, it goes against generally accepted statistical practice. Statisticians do not ordinarily exclude a variable merely because its effect could be ambiguous. McGraw-Hill’s expert, Dr. Addanki, describes the problem: “When I don’t know what the effect of a variable is going to be, to leave it out is to elevate ignorance to arrogance.” (Hrg. Tr. at 77.) Second, McGraw-Hill notes that there are very good reasons to believe that construction-volume data will have a significantly larger effect on the national market than on the local market—namely, that national firms (which, presumably, are the only customers in the market for national CPI) were hit harder by the 2008 recession than were state and local firms. Finally, the Court notes that Warren-Boulton concedes that the price indices are highly negatively correlated with the construction-volume data, indicating that it has an effect and that the effect is significant and negative. These three observations are sufficient to show that construction-volume data is a “major” variable under prevailing case law, and, therefore, its omission is fatal to Dr. Warren-Boulton’s first model. 3. Multicollinearity Dr. Warren-Boulton added construction volume data to his model in response to McGraw-Hill’s contention that it might be causing the result that Dr. Warren-Boul-ton attributes to McGraw-Hill’s misconduct. But when Dr. Warren-Boulton added construction volume data to his model, a new problem arose: multicollinearity. Multicollinearity problems typically arise when the independent variable is correlated with one of the control variables. If the control variables move together with the independent variable, it becomes impossible to isolate the effect of the independent variable on the dependent variable—which, after all, is the goal of regression analysis. The multicollinearity problem manifests itself through low statistical significance of the independent variables. Because of the correlation • between the explanatory variables, there is insufficient variation in the data set to produce statistically significant results. When Warren-Boulton added national construction volume to his regression analysis, he ran into a severe multicollinearity problem. It turns out that construction volume is highly correlated with both the independent and dependent variables. When construction volume is added to the model, the explanatory power of the quarterly time variables plummets. This, McGraw-Hill contends, is fatal to the model. McGraw-Hill argues that the effect captured by Warren-Boulton’s model is really due to the construction-volume data, not McGraw-Hill’s misconduct. Warren-Boulton responds by arguing that the effect cannot be due to the national construction-volume data because when that data is analyzed on its own it produces bizarre results. When considered alone, construction volume has positive effects on McGraw-Hill national prices and negative effects on McGraw-Hill local prices—but it has the reverse effects in both categories for Reed prices. Thus, Warren-Boulton argues, the price effect cannot be explained by construction volume—and that means it is explained by McGraw-Hill’s malfeasance. This explanation alone is insufficient because many other things—for which Dr. Warren-Boulton did not control when he individually assessed the effect of construction volume—could have created this effect. Perhaps, for example, Reed improved its products in a few important markets while McGraw-Hill did not. Further, Warren-Boulton seems to be relying on the assumption that an economic downturn would have a similar effect oh local and national prices for CPI. But one would expect national firms—which are probably the only customers interested in national CPI data—to be hit harder by national downturns than local firms. Similarly, if construction volume is really having no effect—as Warren-Boulton must conclude in order to avoid the consequences of mul-ticollinearity—why is it so well correlated with the dependent variables in the first place? These concerns are significant and contribute to the Court’s conclusion that Dr. Warren-Boulton’s method is inadmissible under Rule 702. 4. Statistical Insignificance McGraw-Hill also objects that Dr. Warren-Boulton’s models are not statistically significant to any reasonable degree and that this alone ought to doom his regression analysis. Specifically, McGraw-Hill contends that Dr. Warren-Boulton’s second model—the one in which he included construction-volume data—produces no statistically significant result for the quarterly time variables. To the extent that McGraw-Hill’s objections on this point center on model two, they are duplicative of its objections based on multicollinearity. The lack of statistical significance in model two arises only because of the addition of construction-volume data, which was the source of the multicollinearity problem. Thus, without the construction-volume data, the analysis does not suffer from a problem of statistical significance. McGraw-Hill’s concerns about multicollinearity are addressed above. And to the extent that McGraw-Hill’s objections center on Dr. Warren-Boulton’s calculation of the price indices themselves, the objections are ■ unavailing. There was some confusion at the Daubert hearing about which model was actually Dr. Warren-Boulton’s “first.” In order to construct the price indices themselves, Dr. Warren-Boulton ran regressions on the pricing data for the purpose of controlling for the quantity and the quality of the data purchased in each CPI subscription. McGraw-Hill argued that these statistics themselves were flawed because Dr. Warren-Boulton did not control for construction-volume data when calculating them. But the price indices focus only on what the prices of the services were, not what caused them to move. To answer that question, Dr. Warren-Boulton needed to control for factors internal to the products themselves. Logically, construction-volume data has no relevance to that question. These objections are not persuasive and do not contribute to the Court’s finding that Dr. Warren-Boulton’s testimony is inadmissible under Rule 702. 5. Pooling McGraw-Hill objects to Warren-Boul-ton’s choice to combine national and local pricing data in his regressions. Combining data in this way is called “pooling.” McGraw-Hill argues that under prevailing statistical methodology, pooling is inappropriate because Warren-Boulton’s own assumptions suggest that national and local prices react differently to the important factors he seeks to measure and because the data indicate that the two markets are different in important ways. Pooling problems arise when data from meaningfully different categories are combined together in a regression analysis. This would be like trying to calculate the effect of location on the price of an apartment by combining New York, Chicago, and Topeka. There are several tests for pooling problems, but many statisticians rely on the Chow test. See, e.g., Gregory C. Chow, Tests of Equality Between Sets of Coefficients in Two Linear Regressions, 28 Econometrica 591, 591-605 (1960). This test uses a mathematically complicated formula to determine whether the independent variables have significantly different effects on two or more subsets of the data. Dr. Warren-Boulton testified that his model unequivocally failed the Chow test under any sensible statistical methodology. But, he claims that the Chow test merely indicates a “safe harbor”: if one’s model passes the Chow test, pooling is safe, but if one’s model does not pass the Chow test, one must appeal to common sense to determine whether pooling is safe. (Hrg. Tr. at 82-83.) To support this reading of the Chow test, Dr. Warren-Boulton points to two scholarly articles in which analysts sought to examine the effect of cigarette pricing on cigarette consumption throughout the United States. Those analysts ultimately decided to pool the data from all fifty states even though the data failed the Chow test. McGraw-Hill counters that these articles are not on point. In the articles, the analysts were not trying to compare national data with state-level data—which is exactly what Dr. Warren-Boulton is trying to do here. McGraw-Hill has persuasively distinguished these articles on that ground. Nonetheless, Warren-Boulton counters this objection by noting that the number of national observations is comparatively low and that the risk of manipulation from pooling the data is small. On his view, national and local data should similarly respond to changes in the control variables, even if they manifest significantly different responses to the independent variable. Similarly, Warren-Boulton claims that the unpooled data lead to “economically nonsensical” results. (Hrg. Tr. at 107.) At the Daubert hearing, Dr. Warren-Boulton said that his decision to pool the data was based on the fact that the unpooled data lead to results that violated his “prior expectations.” (Hr’g Trans, at 87.) McGraw-Hill’s rebuttal expert, on the other hand, ran the regression analysis with the unpooled data and claims simply to have reached the opposite result that Warren-Boulton did—running the regressions on the data sets separately, Dr. Ad-danki claims to have found no damages at all. The issue with Dr. Warren-Boulton’s two responses is that they do not rest on qualities of his expertise. Warren-Boul-ton is, by training, an economist and an econometrician. He is qualified to opine only on areas within the scope of those fields. Ultimately, Dr. Warren-Boulton decided to pool the data because “it made sense to do it this way.” (Dr. Warren-Boulton Deposition Trans. 948:17-22.) He ultimately conceded that his prior expectation is “not sufficient to justify [his] decision to pool.” (Hrg. Tr. at 128.) He similarly conceded that there is no solid statistical reason for his decision to pool the data either (id. at 245), and responded to McGraw-Hill’s objections on the point by noting that “[t]here is no computer program that will let you answer this; it requires someone to look at it and use judgment.” (Id.) This might be true, and there certainly is no reason to believe that Dr. Warren-Boulton’s judgment is anything other than perfectly sensible from a common-sense perspective. The problem is that he cannot offer any methodological explanation for why his judgment came out the way it did. See, e.g., Bricklayers, 752 F.3d at 87-89. And that judgment completely alters the conclusions of the model: had he made a different reasonable judgment, the results would be the opposite of what they were. This is insufficiently “scientific” to be admissible under Rule 702. 6. Robustness The final concern with Dr. Warren-Boulton’s methodology ties in with the penultimate one: it is too manipulable to qualify as “scientific” within the meaning of Rule 702. This concern takes its most persuasive form in its criticism of Dr. Warren-Boulton’s choice of the end dates for his data sets. In creating a benchmark model, Dr. Warren-Boulton must compare the price index ratios to a fixed point in time when, he supposes, the effects of McGraw-Hill’s misconduct will have fully dissipated. Dr. Warren-Boulton concedes that this time frame is, more or less, arbitrary. That alone is not necessarily a fatal problem because, indeed, some judgment is called for in any statistical model and, so long as the model is robust with respect to different choices of arbitrary points, there is no pressing issue. But here the choice of the end-date for the observations in Dr. Warren-Boulton’s analysis has an outcome-determinative effect. Dr. Addanki recreated Dr. Warren-Boulton’s model but changed the end-dates—keeping them well within the range that Dr. Warren-Boulton described as “very conservative”—and found that it yielded no damages. (Compare Defendant’s Exhibit 7, at 31, with Warren Boulton Declaration, at ¶¶ 37-38.) Generally, issues such as the choice of a reasonable time-frame in which to examine data are issues of the expert’s credibility that ought to be decided by the jury. See Bazemore, 478 U.S. at 400, 106 S.Ct. 3000. But where, as here, very minor changes in arbitrarily selected model parameters can entirely alter the model’s conclusions, that model is insufficiently robust to withstand the scrutiny of Rule 702. Scientific conclusions cannot depend upon the arbitrary choice of parameters. Dr. Warren-Boul-ton’s choice of parameters was insufficiently methodological to be admissible under Daubert. E. Conclusion Reed, as the proponent of the expert testimony, must establish by a preponderance of the evidence that Dr. Warren-Boulton’s testimony is admissible under Rule 702. Moore v. Ashland Chem. Inc., 151 F.3d 269, 276 (5th Cir.1998). Given the problems discussed above, Reed has not met this burden. Therefore, McGraw-Hill’s motion to exclude Dr. Warren-Boul-ton’s testimoriy is granted, and the Court will not consider his testimony when evaluating McGraw-Hill’s motion for summary judgment. III. Summary Judgment Standard Summary judgment is appropriate when “there is no genuine dispute as to any material fact and the movant is entitled to judgment as a matter of law.” FED. R. CIV. P. 56. A fact is material if it “might affect the outcome of the suit under the governing law,” Anderson v. Liberty Lob by, Inc., 477 U.S. 242, 248, 106 S.Ct. 2505, 91 L.Ed.2d 202 (1986), and a dispute is genuine if, considering the record as a whole, a rational jury could find in favor of the non-moving party, Matsushita Elec. Indus. Co. v. Zenith Radio Corp., 475 U.S. 574, 587, 106 S.Ct. 1348, 89 L.Ed.2d 538 (1986). The initial burden of a movant on summary judgment is to provide evidence on each element of her claim or defense illustrating her entitlement to relief. Vermont Teddy Bear Co. v. 1-800 Beargram Co., 373 F.3d 241, 244 (2d Cir.2004). If the movant meets this initial burden of production, the non-moving party must then identify specific facts demonstrating a genuine issue for trial. Fed. R. Civ. P. 56(f). The court views all evidence “in the light most favorable to the nonmoving party and draw[s] all reasonable inferences in its favor.” Anderson, 477 U.S. at 250-51, 106 S.Ct. 2505. A motion for summary judgment may be granted only if “no reasonable trier of fact could find in favor of the nonmoving party.” Allen v. Coughlin, 64 F.3d 77, 79 (2d Cir.1995) (citation omitted). But the non-moving party cannot rely upon mere “conclusory statements, conjecture, or speculation” to meet its burden. Kulak v. City of New York, 88 F.3d 63, 71 (2d Cir.1996) (citing Matsushita, 475 U.S. at 587, 106 S.Ct. 1348). IV. Lanham Act Reed has brought claims under Section 43(a) of the Lanham Act, which provides a cause of action against [a]ny person who ... uses in commerce ... any ... false or misleading description ... or misleading representation of fact, which ... in commercial advertising or promotion, misrepresénts the nature, characteristics, qualities, or geographic origin of his or her or another person’s goods, services, or commercial activities. 15 U.S.C. § 1125(a)(1). Two theories of recovery are available to plaintiffs under the Lanham Act. Time Warner Cable, Inc. v. DIRECTV, Inc., 497 F.3d 144, 153 (2d Cir.2007). First, plaintiffs can argue that an advertisement is literally false. See Johnson & Johnson v. GAC Int’l, Inc., 862 E.2d 975, 977 (2d Cir.1988). If an advertisement is literally false, consumer deception is presumed and “the court may grant relief without reference to the advertisement’s actual impact on the buying public.” Time Warner, 497 F.3d at 153 (internal quotation marks and citations omitted). Alternatively, plaintiffs can argue that the advertisement, while literally true (or at least not literally false), is likely to mislead or confuse consumers. Id. To recover under this second theory, plaintiffs must show, using extrinsic evidence, that the misleading statement in fact misled consumers. If the deception was willful, consumer confusion is presumed. Resource Developers, Inc. v. Statue of Liberty-Ellis Island Foundation, 926 F.2d 134, 140 (2d Cir.1991). The burden then shifts to the defendants, who must offer evidence that consumers were not, in fact, confused. Id. Under either theory, plaintiffs must prove that the statement was “material[ ]” to consumers and “involved an inherent ... quality of the product.” Nat’l Basketball Ass’n v. Motorola, Inc., 105 F.3d 841, 855 (2d Cir.1997). They also must prove that that the statement constitutes “advertising and promotion” within the meaning of the Lanham Act. Fashion Boutique of Short Hills, Inc. v. Fendi USA, Inc., 314 F.3d 48, 58 (2d Cir.2002), Reed alleges the following false or misleading statements: • Roper’s Involvement: The six Roper Reports stated that Roper “oversaw the entire comparison process” and “ensured that comparable categories were used in the comparison.” Reed claims that “a jury could conclude ... that Roper’s involvement was a sham.” (Plaintiff’s Memorandum, at 47.) • Roper Report Comparisons: The Roper Reports overstated the number of projects in McGraw Hill’s database as compared to Reed’s database. Specifically, (1) one Roper Report compared Reed’s database to Dataline, one of McGraw-Hill’s other projects, rather than Dodge Network; (2) McGraw Hill excluded some of Reed’s utilities projects in their comparisons; (3) McGraw Hill double-counted some of its own projects; (4) McGraw Hill included its own ASAP projects while excluding Reed’s ASAP projects from the comparisons; and (5) McGraw Hill selected search criteria that were designed to highlight its relative strengths. Reed claims that items 1, 3, and 4 were literally false while 2 and 5 produced misleading project counts. (Defendant’s Memorandum, at 49-50.) • Ad Hoc Comparisons: McGraw Hill provided ad hoc comparisons of search results from McGraw Hill’s and Reed’s databases. Reed alleges that these ad hoc comparisons replicated the methodological flaws in the Roper Reports (detailed above) and thus were literally false. • Stale Executive Briefs: From 2008 through 2012, McGraw Hill published Executive Briefs which printed Roper Report comparisons from 2007 despite the availability of less favorable comparisons from 2008. Reed claims that this staleness conveyed a “false and misleading impression.” (Plaintiffs Memorandum, at 50.) • Claims of Exclusivity: Occasionally, McGraw Hill informed customers that it had exclusive access to projects that were actually also available on Reed Connect. McGraw Hill further claimed that it knew about these exclusives because of customer tips rather than because of its own illegal access. Reed alleges that both the statement of exclusivity and the description of provenance were literally false. • Claimed Project Ratios: McGraw Hill informed customers that it had a 5:1 advantage over Reed’s database in exclusive projects and a 3:1 advantage in all projects. Reed claims that these statements were literally false. A. Advertising As a preliminary matter, the Court must determine whether the challenged statements constitute “advertising and promotion” within the meaning of the Lanham Act. The Second Circuit has adopted a three-part inquiry for determining what constitutes “advertising or promotion.” Fashion Boutique of Short Hills, Inc. v. Fendi USA, Inc., 314 F.3d 48, 58 (2d Cir.2002). Under the Fendi test, “the contested representations must be [1] commercial speech; ... [2] made for the purpose of influencing consumers to buy defendant’s goods or services; and, [3] ... disseminated sufficiently to the relevant purchasing public.” Id. at 56 (quoting and partially adopting the test from Gordon & Breach Sci. Publishers S.A. v. Am. Inst. of Physics, 859 F.Supp. 1521, 1536 (S.D.N.Y.1994) (Sand, J.)). Here, the first two issues are not contested. McGraw Hill argues only that some of the misrepresentations described above, when considered in isolation, were not disseminated widely enough to satisfy Fendi. Before the Court can consider whether the statements were sufficiently distributed, it must answer a preliminary question: are the statements to be considered together or in isolation? The ad hoc comparisons, for example, were distributed to many different customers, sometimes one at a time. Calling a single customer is not “advertising or promotion” within the meaning of Fendi, so if the comparisons are considered one by one, many will not make the cut. McGraw-Hill argues that the Court should look at the statements one by one, while Reed argues that the Court should consider the statements as part of an overall campaign of publicity. Courts assessing Lanham Act claims are to consider an allegedly false “advertisement ... in its entirety,” Avis Rent A Car Sys., Inc. v. Hertz Corp., 782 F.2d 381, 385 (2d Cir.1986) (Friendly, J.), and should “not ... engage in disputatious dissection.” Id. In this case, though, the question is whether a series of individual statements to customers constitutes one “advertisement”—that should not be disputatiously dissected—or a series of advertisements for which dissection is appropriate. “Level of generality is destiny in interpretive disputes.... ” Thomas More Law Ctr. v. Obama, 651 F.3d 529, 560 (6th Cir.2011) (Sutton, J., concurring), abrogated by Nat’l Federation of Independent Business v. Sebelius, — U.S. -, 132 S.Ct. 2566, 183 L.Ed.2d 450 (2012). Here, the Court must decide a sensible level of generality at which to assess the challenged statements. McGraw Hill cites Seven-Up Co. v. Coca-Cola Co., 86 F.3d 1379, 1385 (5th Cir.1996) to support its position that the Court ought to consider each statement separately. There, however, the Fifth Circuit held that “presentation materials ... specifically developed to target ... independent bottlers[] and convince them, based on comparative statistics, to switch from 7UP to Sprite,” id, were “advertising” for Lanham Act purposes. Coca-Cola had presented the materials to eleven bottlers in separate presentations; the presentation in its entirety may only have been shown to two bottlers. Nonetheless, the “presentation” was advertising. The court appeared to consider the presentation as a whole and evaluated the total number of people to whom it was presented. Id. at 1386. This case, therefore, does not support McGraw-Hill’s piecemeal approach. Similarly, in Gordon & Breach—an influential opinion relied upon by several circuit courts in formulating their tests for what constitutes advertisement—Judge Sand described an allegedly misleading advertising and promotional campaign comprising several elements. 859 F.Supp. 1521. As in this case, the plaintiff alleged that the defendant published a misleading comparison between two competitors and falsely claimed that it was ah independent survey. In addition to the survey itself, the plaintiff challenged collateral uses of the survey results, such as a press release, a letter to the editor of a relevant trade publication, and individual comments to customers. After identifying the various elements of the campaign, the court distinguished between those elements that were properly deemed “commercial” speech and those elements that warranted heightened First Amendment protection as “protected editorial speech on a topic of public significance.” Id. at 1544. While the court separated constitutionally distinct categories of speech, it did not advocate further atomizing elements of the promotional campaign. Instead, on the matter of distribution, the court cited the oft-repeated refrain that “the level of circulation required to constitute advertising and promotion will undeniably vary from industry to industry and from case to case.” Id. at 1535 (internal quotations marks and citations omitted). It did not distinguish between the individual instances of allegedly false communication. Neither will this Court. Therefore, the question is whether, taken as a whole, McGraw-Hill’s efforts constituted “advertising” within the meaning of the Lanham Act. The Act’s “reach is broader than merely the classic advertising campaign.” Gordon & Breach, 859 F.Supp. at 1534 (internal quotation marks and citations omitted). But it does not extend to “statements in oral conversations by an individual sales representative to an individual customer.... ” Licata & Co. v. Goldberg, 812 F.Supp. 403, 408 (S.D.N.Y.1993). Here, McGraw-Hill tried to convince customers of the superiority of its service through a series of individual conversations. Unlike the individual conversations in Licata, the ad hoc comparisons at issue in this case were an undisputed part of a broader campaign to compete with Reed and to tout the supposed advantages of the Dodge Network over Reed Connect. In that context, what would ordinarily seem like individual conversations take on added significance. There is evidence that McGraw-Hill management directed individual salespeople to disseminate several of the allegedly false or misleading statements. (E.g., Plaintiffs Exhibit’s 512-518.) There is little difference between this and a traditional advertising campaign in either purpose or effect. The purpose is to win customers from a competitor on a large scale and the effect—assuming there is one—is the same. As in Gordon & Breach, the mere fact that the promotional campaign took the form of individual conversations does not mean that it is not advertising when taken as a whole. Taken together, McGraw-Hill’s statements are advertising within the meaning of the Lan-ham Act. B. Falsity As previously mentioned, a statement can be “false” for Lanham Act purposes in two ways: it can be “literally false, i.e., false on its face,” or it can be implicitly false because it is “misleading].” DIRECTV, 497 F.3d at 153. The difference between the two classes of falsity is the comparison that each invites. A determination of literal falsity invites a comparison “between the statement and ... reality,” while a determination of implicit falsity invites a comparison between the “impression on the listener” and reality. Id. (quoting Sobering Corp. v. Pfizer Inc., 189 F.3d 218, 229 (2d Cir.1999)). Statements are to be read in context, id. at 158, which has led the Second Circuit to adopt what other circuits have called the “false by necessary implication” doctrine. Id. (citing, inter alia, Scotts Co. v. Unit