By Alexander M. Waksman
Imagine a scientist who has total conviction that a particular hypothesis is correct. The scientist’s belief in their hypothesis is so strong that they dispense with the need to test it in experiments or compare it against empirical data. Even when third parties carry out such tests, the scientist considers it unnecessary to take account of the results. We would condemn that approach as “bad science”.
Antitrust may not be a science in the strict sense, but the same logic applies. When a competition authority (or claimant) identifies a theory of harm, that theory ought to be tested against actual market developments to determine whether its predictions are correct. As Commission guidance states, “economic theory is used to develop a testable hypothesis that is later checked against the data. In that case, the economic analysis makes predictions about reality that can be tested by observations and potentially rejected or verified.”
Thus, a theory of harm is useful as a starting point, but it cannot be the whole story. In particular, when the challenged conduct has been ongoing for a significant period of time and sufficient data are available, it makes good sense to ask whether the hypothesised anticompetitive outcomes have actually transpired.
The need to “test the hypothesis” is a controversial topic in abuse of dominance cases, where competition authorities, courts, and commentators offer differing formulations of the applicable legal standard. It has particularly important consequences for firms operating in online markets, which are drawing increased antitrust intervention.
- The Legal Standard: “Capability” or “Likelihood” of Restricting Competition
Former Commissioner for Competition, Mario Monti, observed that “in competition the best should win on the merits, but only on the merits.” The central challenge in abuse of dominance cases is therefore to distinguish competition on the merits from anticompetitive exclusionary conduct. This is a difficult distinction to draw, particularly in online markets, where new business models, parameters of competition, and monetisation strategies frequently emerge.
One of the conditions for determining whether conduct is abusive is whether the conduct has the “effect” of excluding rivals. Otherwise, as Advocate General Wahl observed, there would be a risk that “competition law sanctions form.” The precise meaning of “anticompetitive effects” has not been interpreted consistently, however.
On one view, conduct needs merely to be “capable” of restricting competition. Advocates General Wahl and Kokott have interpreted this as denoting conduct that is “likely” to restrict competition – albeit with some disagreement as to the level of “likelihood” that is required. The Court of Justice has stated that “only dominant undertakings whose conduct is likely to have an anticompetitive effect on the market fall within the scope” of abuse of dominance rules, and that “such an assessment seeks to determine whether the conduct of the dominant undertaking produces an actual or likely exclusionary effect, to the detriment of competition” (emphasis added).
Beyond the distinction between “capability” and “likelihood” (or lack thereof), the Court’s formulation raises a broader question: when must competition authorities take into account “actual” effects and when is it sufficient only to show a “likelihood” or “capability” of restricting competition?
Traditionally, the European Commission and Courts treated “actual” and “likely” effects as alternatives. In other words, the requirement to show exclusionary effects could be satisfied by demonstrating either that the conduct did in fact exclude competitors (the “actual” effect) or that it was likely to exclude rivals in the circumstances of the case (the “likely” effect). While recognising that anticompetitive effects could not be merely “hypothetical”, the Court held that there was no need to show that the “likely” impact materialised into “concrete” or “actual” exclusion.
But the distinction between “actual” and “likely” effects is not always clear-cut. If the alleged “likely” effects do not occur, that casts doubt on whether such effects were ever truly “likely”. In such circumstances, it is for the party alleging an abuse to explain why the theory of harm is correct, even though its predictions have not been borne out in practice. And in a series of abuse of dominance cases, the European courts have treated actual effects as having significant probative value.
2. The Relevance of Actual Exclusionary Effects
In Microsoft the Commission determined that a tie between Windows OS and Windows Media Player could not be assumed to foreclose competition without further analysis. It undertook a detailed assessment of the actual effects of the tie, including a review of its impact on the number of subscribers, users, and usage shares of rival media players over the course of five years. As the General Court observed, “the Commission examined the actual effects which the bundling had already had on the market.” The Commission took the same approach in the Microsoft (Tying) case, concerning the tie between Microsoft’s Windows operating system and the Internet Explorer browser. The commitments decision noted that the Commission “had examined such effects closely,” including through extensive empirical surveys.
The Microsoft cases thereby established a need to look closely at actual market developments in circumstances where the conduct cannot be considered “by nature” abusive. Summarising the Windows Media Player case, Briss J stated that “the Commission [in Microsoft] had examined the actual effects on the market… For abuse to be established there must be a finding that the practice in question forecloses competition. The legal principle I take from the authorities is simply that such a finding may be based on inference but the inference must be justified in all the circumstances. Just because it is normally assumed, it does not follow that it will always be assumed. The circumstances may be such that such an assumption cannot be made and a close analysis of the actual effects is required.”
Likewise, in Posten Norge the EFTA Court accepted that “in some cases, the persistent lack of actual negative effects on competition may cast doubt on a finding by ESA that a certain conduct is liable to restrict competition.” On the facts of the case, though, the actual effects left “no doubt” that Posten Norge’s conduct had impeded competition. Specifically, the Court had “no doubt that the applicant’s conduct made entry for its competitors more difficult”, including because of the existence of actual exclusionary effects: Posten Norge’s competitors “were unsuccessful in securing cooperation with Norway Post’s partners while the exclusivity agreements were in force, but succeeded in achieving such cooperation after the conduct had ended.” 
In Post Danmark I the Court again treated actual exclusionary effects as relevant to the analysis, although this time as evidence that the conduct was not abusive. In reviewing the allegedly unlawful pricing strategy, the Court observed that the complainant had not been excluded or marginalised. Rather, it had “managed to maintain its distribution network despite losing the volume of mail related to the three customers involved and managed, in 2007, to win back” two of these customers – an outcome that was not redolent of exclusion from the market.
The approach in Microsoft, Posten Norge, and Post Danmark II is consistent with developments in the case law concerning Article 101, where – outside the narrow group of “by object” infringements – an agreement must be shown to have actual effects on the market for the agreements to be considered anticompetitive. There is no particular reason why the effects analysis should be different with respect to unilateral conduct.
3. Clarifying The Principles: Streetmap v Google
The relevance of actual effects was addressed in the recent Streetmap v Google case, which concerned a Google Maps thumbnail being placed at the top of the Google search results page. As an initial matter, Roth J noted that the Commission has – in many cases – examined the “experience over the years since the abuse started”, particularly in cases where the conduct is not one of the “classic forms of abuse”, citing Microsoft.
Roth J went on to state that “in a case such as the present, I would find it difficult in practical terms to reconcile a finding that conduct had no anti-competitive effect at all with a conclusion that it was nonetheless reasonably likely to have such an effect… the appropriate approach, it seems to me, is that it is for Streetmap to establish that the conduct was reasonably likely to harm competition. In determining that question, the court will take into account, as a very relevant consideration, evidence as to what the actual effect of the conduct has been.”
This approach follows an intuitive logic: If the conduct has failed to produce actual exclusionary effects, that casts doubts on whether it was ever likely to do so. And although Roth J did not explain what he meant by a “case such as the present”, the CAT referred subsequently to the fact that the conduct in Streetmap had been ongoing for eight years and concerned internet searches – an activity that occurs “multiple times each day”. Thus, “if [the conduct] were to have an anti-competitive or foreclosure effect, it was inconceivable that this would not have become evident.”
Having carried out the assessment, the judgment concluded that introducing the Google Maps thumbnail did not have an appreciable effect in taking away custom from Streetmap. Roth J stated: “In the light of that, I conclude that it was not reasonably likely to give rise to anti-competitive foreclosure.” In other words, since the conduct at issue did not have an actual appreciable effect, it could not be described as likely to produce anticompetitive foreclosure.
The Streetmap judgment is important not only as a matter of legal principle, but it could also have an impact on enforcement in other digital markets, which are drawing increased attention from antitrust authorities. The European Commission has carried out investigations into Google Shopping, the Android operating system, AdSense for Search, and Amazon’s eBooks contracts. The French and German competition authorities have outlined putative theories of harm relating to the exclusionary use of “big data” by dominant companies – an area which is also being probed in a national sector inquiry in Italy. And the UK competition authority has confirmed online markets as a “policy priority.” It is therefore worth considering what the cases mean for the effects analysis in online markets specifically.
4. Three Conditions For Assessing Actual Effects
The foregoing shows that actual effects can be highly relevant – or even dispositive – as to the question of whether exclusion was “likely” or “capable of” occurring. The relevance of actual effects will vary on a case-by-case basis, though, depending on the following factors that can be discerned from the case law.
Factor 1: Whether the conduct is “by nature” liable to restrict competition. Conduct that cannot be presumed to have anticompetitive effects, in the circumstances or market context of a given case, needs to be examined closely by reference to its actual effects on competition (per Microsoft and Huawei v Unwired Planet). Thus, conduct that is “by nature” abusive may be treated as infringing Article 102 TFEU (subject to any objective justification) without recourse to a full analysis of actual effects – albeit the scope of “by nature” abuses is diminished, if not eliminated, following the recent Intel judgment). By contrast, conduct that is not inherently abusive may require an assessment of actual market developments (depending on Factors 2 and 3, below).
Factor 2: Whether exclusionary effects would have been felt at the time of assessment. In some cases, proceedings may arise at an early stage, too soon for the alleged anticompetitive effects to have manifested themselves. Where intervention occurs at a later stage, though, it may be the case that any exclusionary effects would have been felt at the time when the conduct is assessed. In the latter circumstance, actual market developments should be taken into account as part of the competitive assessment (per Streetmap v Google).
This factor depends on how long the conduct was in place (or how long ago it ceased), and whether there was a significant number of transactions during the period of the alleged abuse (per Socrates v Law Society). In essence, the question is whether any anticompetitive effects would have occurred by the time of carrying out the antitrust assessment.
In previous cases concerning online markets, the challenged conduct has been in place long enough to cover a large number of interactions with customers or consumers, such that any anticompetitive impact would have been felt. In Streetmap v Google, the conduct lasted for eight years, during which time users entered a vast number of search queries. If the conduct was “likely” to affect the position of rivals, that would have been evident at the time when the High Court heard Streetmap’s case. Likewise, the Microsoft cases concerned conduct that lasted several years.
Even when conduct is in place for a short period of time, though, or has only recently begun, the sheer volume and frequency of user interactions with online services mean that any effects may have already been felt. Indeed, this is inherent in the way that online service providers carry out research into their own products. For example, in 2012 Facebook conducted a study to test whether changing users’ news feeds to show more “positive” or “negative” items could affect the emotions expressed in a user’s own posts. Despite the experiment lasting just one week, researchers were able to benefit from a very large sample size of 689,003 users and could analyse over 3 million posts, resulting in empirical support for the far-reaching claim that “emotions can spread throughout a network.” This experiment confirms that conduct may trigger a change in observable user behaviour in even a short space of time.
Factor 3: Whether data concerning effects are available. Although not discussed in the case law expressly, a further factor may be the availability – or ability to collect – data concerning actual effects, without which a review of actual effects would be impossible. Commission guidance on the submission of economic evidence notes that the limitations of a dataset should be recognised, although deficiencies in the data should not necessarily prevent an economic analysis from carrying weight.
The amount of user interactions with online services will in many cases generate sufficient data for actual effects to be examined. By way of illustration, Ofcom reported that the average adult Internet user in the UK spends over 20 hours online per week, as of 2015. By August 2016, the average adult was reported to be spending almost 9 hours each day on media and communications activities – more time than spent sleeping. In July 2017, consumers in Britain spent on average £1.1 billion online each week – an increase of 15.1% compared with the previous year. And in the Google Shopping case the Commission reports having analysed 5.2 Terabytes of search results data, equating to 1.7 billion search queries.
5. “Bad Science” In Online Markets
In competition law – as in science – the null hypothesis is true until proven otherwise. If a theory of harm predicts the exclusion or marginalisation of a dominant company’s rivals, only a “bad scientist” would disregard evidence that these predictions proved incorrect. In online markets particularly, it is important not to assume that a theory of harm is correct without testing the theory’s predictions against real world events. That is because conduct that may be liable to exclude rivals in offline markets could have a very different outcome in an online environment.
This was the insight of the Commission decision in Microsoft, where the ability of users to download alternatives to Windows Media Player (the tied product) – sometimes for free – meant that anticompetitive effects could not be presumed, and an assessment of actual effects was needed. The Commission’s recent decisional practice in merger cases confirms the need to assess the specific features of online markets, rather than extrapolating from the offline environment, given features such as low barriers to entry, multi-homing, and ease of switching between services online.
In Microsoft/Skype, the Commission considered that Microsoft would lack incentives to tie or bundle its products with Skype following the concentration. In particular, the Commission noted that consumers would not simply use “whatever communications product is provided with Windows” – rather, consumers use multiple communications services across different platforms. Skype was offered for free, and consumers would likely migrate to other free services if Microsoft tried to impose a charge. Switching was particularly easy since rival services were provided directly on the Internet, avoiding even the need for user downloads.
Similar insights are set out in the Commission’s decision in Facebook/WhatsApp, which found “no significant costs preventing consumers from switching between different consumer communications apps”. The services were offered for free, were easily downloadable, take up little memory space and are easy to use, and consumers could switch between them with ease. Therefore, there were no “significant ‘traditional’ barriers” to entry.
A review of actual effects therefore serves to avoid false positives in antitrust enforcement – the so-called “Type I error”. In a recent speech concerning antitrust enforcement in the digital sector, FTC Commissioner Ohlhausen commented that “regulators are just people entrusted, temporarily, with the levers of government power. We possess no crystal ball with which we can unerringly see the road ahead, and we can be just as wrong about the way the future will unfold as anyone else. In my time here at the FTC, and throughout my career, I have tried to stress the need for regulators to acknowledge the limits of their actual knowledge before exercising government power.” This seems particularly true of the fast moving and unpredictable online world.
The foregoing analysis should not be taken as meaning that competition authorities (or claimants) are required to show actual exclusionary effects in every case. As commentators have pointed out, such a position would require authorities to wait for observable anticompetitive effects to arise before intervening, by which point it may be too late to restore effective competition.
Rather, the need to examine actual effects depends on the interplay of the three factors listed above: whether there is a “by nature” abuse; whether actual effects would already have been felt; and whether sufficient data to measure those effects are available. In digital markets, this will more often than not mean that actual effects are highly relevant – or even dispositive – to the antitrust assessment. Relying on a hypothesis that conduct is “likely” or capable” of excluding rivals is not enough.
 The author is an associate at the London office of Cleary Gottlieb Steen & Hamilton LLP. This article reflects the views of the author. It does not claim to represent the views of Cleary Gottlieb or its clients. The author is grateful for the comments of Nadja Huber on an earlier draft. Any errors or omissions are the sole responsibility of the author.
 Commission Guidance On Best Practices For The Submission Of Economic Evidence And Data Collection In Cases Concerning The Application Of Articles 101 and 102 TFEU And In Merger Cases (2010), paragraph 13.
 Mario Monti, Speech at the conference ‘Antitrust in a Transatlantic Context,’ Brussels, June 7, 2004.
 Opinion of AG Wahl, Case C-413/14 P Intel v Commission ECLI:EU:C:2016:788, paragraph 118. This article is concerned with exclusionary conduct. It does not address “exploitative” forms of abuse, for which a different analytical framework applies. The other conditions include (1) that the conduct has a “form” that creates the possibility of using market power to exclude rivals (e.g., tying a dominant product with non-dominant one), and (2) the conduct is not objectively justified.
 Case C‑413/14 P Intel v Commission ECLI:EU:C:2017:632, paragraph 138-140.
 Opinion of AG Kokott, Case C-23/14 Post Danmark II ECLI:EU:C:2015:343, paragraph 82 (“According to settled case-law, it is necessary but also sufficient that the rebates in question can produce an exclusionary effect. (54) This is the case where, on the basis of an overall assessment of all the relevant circumstances of the individual case, the presence of the exclusionary effect appears more likely than its absence”); and Opinion of AG Wahl, Case C-413/14 P Intel v Commission ECLI:EU:C:2016:788, paragraph 117 (“The aim of the assessment of capability is to ascertain whether, in all likelihood, the impugned conduct has an anticompetitive foreclosure effect. For that reason, likelihood must be considerably more than a mere possibility that certain behaviour may restrict competition. Contrariwise, the fact that an exclusionary effect appears more likely than not is simply not enough”).
 Case C‑23/14 Post Danmark II ECLI:EU:C:2015:651, paragraphs 67 and 69. See also C‑209/10 Post Danmark I ECLI:EU:C:2012:172, paragraph 44.
 See e.g., Case C‑23/14 Post Danmark II ECLI:EU:C:2015:651, paragraph 66; and Case C‑52/09 TeliaSonera Sverige EU:C:2012:172, paragraph 44. See also M. Kadar, The Meaning of “Anticompetitive Effects” Under Article 102 TFEU, CPI Antitrust Chronicle, March 2016 (the Courts have elaborated a standard “to identify a middle ground between purely hypothetical effects and actual effects”).
 Case COMP/AT.37792 Microsoft (Windows Media Player), Commission decision of March 24, 2004, paragraphs 841 and 905-926.
 Case T-201/04 Microsoft ECLI:EU:T:2007:289, paragraph 1035.
 Case COMP/AT.39530 Microsoft (Tying), Commission decision of December 16, 2012, paragraph 34. The Commission’s assessment involved large scale empirical studies, such as “empirical surveys of the actual web browser usage characteristics of both consumers and enterprises” (footnote 23 and paragraphs 50-54).
 Unwired Planet v Huawei  EWHC 711 (Pat), paragraphs 529-530. On the facts of the case, Briss J held that, given the prevalence of the conduct at issue (worldwide licensing of patent families) “I am not prepared to assume that the tying of a SEP licence in one country to a SEP licence in another country has by its nature a competitive foreclosure effect. A close analysis of the actual effects would be required and that has not been done” (paragraph 550).
 Case E-15/10 Posten Norge AS v EFTA Surveillance Authority, judgment of April 12, 2012, paragraph 192.
 Case E-15/10 Posten Norge AS v EFTA Surveillance Authority, judgment of April 12, 2012, paragraphs 193 and 195.
 Case C-209/10 Post Danmark ECLI:EU:C:2012:172, paragraph 39.
 See Case C‑67/13 P Cartes Bancaires ECLI:EU:C:2014:2204, paragraphs 51-52, 58; and Case C-345/14 Maxima Latvija ECLI:EU:C:2015:784, paragraphs 18-19.
 Streetmap v Google  EWHC 253 (Ch), paragraph 89.
 Streetmap v Google  EWHC 253 (Ch), paragraph 90.
 Socrates Training Limited v. The Law Society of England and Wales  CAT 10, paragraph 149. The same was not true of the conduct in Socrates, which was ongoing, had only recently begun, and affected transactions that took place only once a year (or even once every two years). That said “this does not mean that it is inappropriate to consider the actual effect on an independent provider like Socrates, but it is necessary to bear in mind that the degree to which conduct has actually had (or not had) an anti-competitive effect is only evidential” (paragraphs 149 to 150).
 Streetmap v Google  EWHC 253 (Ch), paragraph 139. See also Robert O’Donoghue and Jorge Padilla, The Law and Economics of Article 102 (2nd Edition), p.269 (“if conduct has continued for some time with no perceptible effects, then it is legitimate to wonder whether it has any material anticompetitive effect”).
 Case COMP/AT.39740 Google Search (Shopping), Commission decision of June 27, 2017.
 Case COMP/AT.40099 Google Android.
 Case COMP/AT.40411 Google Search (AdSense).
 Case COMP/AT.40153 E-books MFNs and Related Matters (Amazon), Commission decision of May 4, 2017.
 See Bundeskartellamt and Authorité de la Concurrence, Competition Law and Big Data, May 10, 2016, pp.17-25; and Autorità Garante della Concorrenza e del Mercato, “Big Data”: Italian Regulators open a sector inquiry, Press Release, June 1, 2017.
 MLex, Digital market probes are ‘policy priority’ for UK antitrust authority, official says, September 14, 2017.
 Hence in Microsoft (Tying) the Commission stated that the commitments would enable rivals to be the sole, default web browser, which had “not been possible in previous versions of Windows for over a decade”: Commission memorandum, Antitrust: Commission accepts Microsoft commitments to give users browser choice – frequently asked questions, MEMO/09/558, December 19, 2009.
 Adam D. I. Kramer, Jamie E. Guillory, and Jeffrey T. Hancock, Experimental evidence of massive-scale emotional contagion through social networks, Proceedings of the National Academy of Sciences, June 17, 2014, vol. 111, no. 24, pp.8788-8790.
 Commission Guidance On Best Practices For The Submission Of Economic Evidence And Data Collection In Cases Concerning The Application Of Articles 101 and 102 TFEU And In Merger Cases (2010), paragraph 22. In Streetmap v Google, the court was able to assess – and make a finding on – the (lack of) actual effects, even though “this is not an easy assessment due to the limitations in the data” (paragraph 139).
 Ofcom, Adults’ Media Use And Attitudes Report, June 2017, available at https://www.ofcom.org.uk/__data/assets/pdf_file/0020/102755/adults-media-use-attitudes-2017.pdf.
 Ofcom, Communications Market Report, August 2016, available at https://www.ofcom.org.uk/__data/assets/pdf_file/0024/26826/cmr_uk_2016.pdf.
 Office for National Statistics, Retail sales in Great Britain: July 2017, Section 8, available at https://www.ons.gov.uk/businessindustryandtrade/retailindustry/bulletins/retailsales/july2017.
 Commission Press Release, Commission fines Google €2.42 billion for abusing dominance as search engine by giving illegal advantage to own comparison shopping service, Brussels, June 27, 2017. Moreover, Established economic techniques enable competition authorities to use the data to test the presence – or absence – of actual effects. A commonly used technique, for example, is the “difference-in-difference” analysis: see Shawn W. Ulrick and Seth B. Sacher, Inferring Anticompetitive Price Effects From Difference-In-Difference Analysis: A Caveat, Journal of Competition Law & Economics, Volume 11, Issue 4, December 1, 2015, pp.999-1012.
 Paco Garcia-Gonzalez, Explainer: what is a null hypothesis?, The Conversation, December 20, 2012, available at http://theconversation.com/explainer-what-is-a-null-hypothesis-10757 (“Hypotheses developers and testers usually hope that the null hypothesis is rejected and their alternative hypothesis supported – that the drug they’re testing is effective; the campaign they’re running is a success; that the light is bent by gravity as predicted by Newtonian physics and Einstein’s theory of relativity… It might seem obvious, but until proven otherwise, the null hypothesis is true”).
 Case COMP/M.6281 Microsoft/Skype, Commission decision of October 7, 2011, paragraphs 152, 157, 167.
 Case COMP/M.7217 Facebook/WhatsApp, Commission decision of October 3, 2014, paragraphs 109-110 and 117.
 Maureen K. Ohlhausen, Antitrust Enforcement in the Digital Age, Global Antitrust Enforcement Symposium, September 12, 2017, p.5.
 See e.g., Robert O’Donoghue and Jorge Padilla, The Law and Economics of Article 102 (2nd Edition), p.269.