Readers of Truth on the Market are no doubt aware of Judge Amit Mehta’s Aug. 5 decision in the Google search antitrust case—that is, his 286-page memorandum and order finding Google liable for violating Section 2 of the Sherman Act (specifically, illegal monopoly maintenance in two markets: general search services and general text advertising).
Comments have been coming fast and furious (see, for example, Alden Abbott here; Greg Werden here, and in July, in advance of the opinion, here; an interesting thread by Herb Hovenkamp here; Asheesh Agarwal here, a thoughtful and more detailed International Center for Law & Economics white paper from Geoff Manne here; a post from Josh Wright here; and there’s a nice (if brief) pre-decision discussion between Randy Picker and Spencer Weber Waller here). That’s hardly surprising, if just the tip of the iceberg. And all await, with barely bated breath, the court’s remedy.
A great deal of interesting commentary preceded the decision. I won’t survey it here, but I do recommend a 2020 article by Judge Douglas Ginsburg and Korin Wong-Ervin. It’s not about the Google case specifically—it’s about challenging consummated mergers under Section 2—but it has a good deal to say about the U.S. Court of Appeals for the D.C. Circuit’s 2001 per curiam opinion in Microsoft, a decision that figures heavily in the Google case. Judge Ginsburg was chief judge of the D.C. Circuit court in 2001, and he may well have authored the Microsoft opinion (although I don’t know that he did).
I’m going to focus on a small technical issue of relevance to the case: really, just a few behavioral-science experiments and the expert testimony about them that seemed so persuasive to Judge Mehta. Something of a spoiler alert: I think that the evidence was misrepresented—oversold—and didn’t show much of anything of relevance to the decision.
But to get there, a bit of background about the case:
Judge Mehta’s opinion is careful, thoughtful and, in many respects, defensible. I also think that he got it wrong on some key points that may find traction on appeal. And there’s much of interest in the opinion. Here, however, I want to focus on the key issue of causation, and a piece of the case that’s been noticed but not so much discussed; that is, the expert testimony on the “stickiness” of browser defaults.
To back up just a bit, Google, unquestionably, has a very large share of general search in the United States. Judge Mehta’s decision says that “[b]y 2020, it was nearly 90%, and even higher on mobile devices at almost 95%.” Maybe that’s about right, but there’s nothing inherently wrong with it from an antitrust perspective. As the D.C. Circuit noted in Microsoft, “merely possessing monopoly power is not itself an antitrust violation.” What was alleged, and what Judge Mehta found, was illegal monopolization; that is, “the willful acquisition or maintenance of that power as distinguished from growth or development as a consequence of a superior product, business acumen, or historic accident.”
Specifically, the U.S. Justice Department (DOJ) alleged, and Judge Mehta found, certain contracts unlawful: Google has entered into agreements with certain device manufacturers (notably, but not exclusively, Apple) and Mozilla, which developed and markets Firefox. Under those contracts, Google is the default general search engine (GSE) pre-loaded in, e.g., Apple’s Safari (and hence on iPhones) and Mozilla’s Firefox browsers. These are deemed de facto exclusive-dealing contracts, although nothing in the contracts precludes access to other browsers or other search engines and it is, in fact, possible for users to switch—some might say it’s easy. And as such, these contracts were alleged to foreclose access to the market by actual and would-be competitors. Access to the markets, that is—access to competition for a certain share of general search and, hence, allegedly, access to the advertising markets at issue.
“Exclusionary conduct” then, was the issue, just as the D.C. Circuit ruled that exclusionary conduct was critical to illegal monopoly maintenance in Microsoft. But the key here is the question of causation. Judge Mehta recognizes that even truly exclusionary contracts are not per se illegal (page 219 of the decision, for those following along). The question is whether the default agreements caused Microsoft and others to be excluded from access to the market, and what DOJ had to show to establish causation.
And that’s where Judge Mehta’s thoughtful and mostly clear decision becomes unclear. Or, as my ICLE colleague Geoff Manne argues (here), Judge Mehta misread, and misapplied, the holding in Microsoft.
Judge Mehta gets the first part of exclusion right. Quoting Microsoft, he points out that, “to be condemned as exclusionary, a monopolist’s act must have an ‘anticompetitive effect.’ That is, the monopolist must harm the competitive process and thereby harm consumers. In contrast, harm to one or more competitors will not suffice.” As the D.C. Circuit had it, “the plaintiff, on whom the burden of proof of course rests . . . must demonstrate that the monopolist’s conduct indeed has the requisite anticompetitive effect.”
But if that’s right, where does he go wrong? Quoting Microsoft again, it’s here:
causation does not require but-for proof. The plaintiff is not required to show that but for the defendant’s exclusionary conduct the anticompetitive effects would not have followed. Such a standard would create substantial proof problems, as “neither plaintiffs nor the court can confidently reconstruct… a world absent the defendant’s exclusionary conduct.”
Now, that’s puzzling. For conduct to have an “anticompetitive effect” is, in effect, to say that the conduct caused the competitive harm at-issue. And what is it to show that causal relationship without showing that, but for the defendant’s exclusionary conduct [or some other intervening cause], the anticompetitive effects would not have followed?
As Manne explains—as do, not incidentally, Ginsburg and Wong-Ervin—the facts in Microsoft were deemed to demand a special (and exceptionally lower) standard of proof. For “nascent competitors,” such as Netscape, the Microsoft court reasoned that plaintiffs could not, as a practical matter, establish the “but for” counterfactual. The circuit court called this an “underlying proof problem.”
But the standard wasn’t lowered simply because the government couldn’t meet a higher one. Numerous findings of fact in Microsoft were taken to have established the effects of Microsoft’s conduct; these, in turn, were deemed to have kept Netscape and others, as matters of fact, from reaching the minimum efficient scale at which they could be competitors to Microsoft. Under those conditions, faced with uncertainty about what Netscape might have become as a competitor, the circuit court reasoned that it could infer causation under a lower evidentiary standard. As Ginsburg and Wong-Ervin review, quoting Microsoft:
“To some degree, ‘the defendant is made to suffer the uncertain consequences of its own undesirable conduct,’” i.e., the practices that had been shown to have had anticompetitive effects. In these carefully limited circumstances, the questions were “(1) whether as a general matter the exclusion of nascent threats is the type of conduct that is reasonably capable of contributing significantly to a defendant’s continued monopoly power and (2) whether Java and Navigator reasonably constituted nascent threats at the time Microsoft engaged in the anticompetitive conduct at issue.” (internal citations omitted)
They note, further, that the point was driven home by the D.C. Circuit’s subsequent decision in Rambus Inc. v. FTC, where “the court held that the agency failed to prove that “but-for” the defendant’s conduct, there would have been harm to the competitive process.”
But Judge Mehta does not think that Google acquired its market share by dint of anticompetitive conduct: “Google has not achieved market dominance by happenstance. It has hired thousands of highly skilled engineers, innovated consistently, and made shrewd business decisions. The result is the industry’s highest quality search engine, which has earned Google the trust of hundreds of millions of daily users.” And he’s right about that. The default agreements did not precede Google’s rise to preeminence—they followed it.
Here, Mehta’s opinion gets a bit confusing. He seems to misread Rambus, reasoning that “Rambus does not establish a categorical rule that the anticompetitive effects of an exclusive agreement must be measured against a but-for world,” because otherwise the 2008 Rambus decision would have repudiated Microsoft, and it doesn’t. No. Microsoft recognized a default rule and a special case; Rambus, decided by the same court seven years later, didn’t fit the special case, so it used (and restated) the default rule. A default rule is not categorical–fair enough–it is defeasible. But it is presumed to hold, absent the conditions of the special case. Under typical conditions, it is the rule.
Does he find that Microsoft’s Bing is a nascent competitor to Alphabet’s Google? Not expressly, no. He doesn’t think he needs to, given his reading of Microsoft. And how could he? It is an established, highly developed, and very well-funded GSE and it’s developed considerably since its entry in 2009. Its annual ad revenues have been reported to be more than $20 billion.
On the other hand, at times, he seems to:
Google’s exclusive agreements have a second important anticompetitive effect: They deny rivals access to user queries, or scale, needed to effectively compete. Scale is the essential raw material for building, improving, and sustaining a GSE.
Whether Judge Mehta misread Microsoft—and, for that matter, Rambus—is a question of law. Questions of law are the most likely candidates for appeal, and will likely receive the lion’s share of commentary. But questions of fact are critical to the decision under either the special-case “edentulous” (toothless—had to look it up the first time myself) standard or the more general one recognized in Rambus.
On some key questions of fact, Judge Mehta’s opinion seems to credit expert testimony from Stanford’s Antonio Rangel, who holds the title Bing [yep] Professor of Neuroscience, Behavioral Biology, and Economics at the California Institute of Technology. As I revealed in a rare confessional moment last November, “[t]his has never been precisely my field, but in my misspent youth (ok, my late 30s), I did some work at the National Institutes of Health (NIH), in the Laboratory of Neuropsychology (LN), so I thought I’d take a look at the underlying research.”
The DOJ has posted Rangel’s slides, and the high-level message is clear: “Search engine defaults generate a sizable and robust bias towards the default” and “[s]earch engine default effects have stronger effects on mobile devices than on personal computers.”
On one reading, that’s unsurprising. Changing anything—changing lunch reservations or seats around a table—implies transition costs, however high or low (or even minuscule), so any sort of default might generate a bias toward the default position, whether that’s anything to do with general search engines or not.
How sizable? Do we know? Rangel’s slides include a few cherry-picked examples that imply large effects, indeed. For example, “In Austria, where citizens were registered as organ donors by default, 99% were registered donors. In neighboring Germany, where citizens had to affirmatively register, only 12% were registered donors.” That might ring a bell with readers of Richard Thaler and Cass Sunstein’s “Nudge.”
Still, one might wonder about the quality or applicability of the evidence, and whether the expert scientist provided a reasonable overview of it. Off the bat—that’s a striking correlation, but it’s just a correlation, and one that may or may not have anything to do with search engine defaults. And one might be unaware of, for example, similar data finding variable and rather lower numbers across EU member states, such that, for example, “presumed consent countries have 28% to 32% higher cadaveric donation and 27% to 31% higher kidney transplant rates in comparison to informed consent countries, after accounting for potential confounding factors.” Still just correlations, but much smaller numbers than we see in the Austria vs. Germany comparison. And variation across EU states on organ-donation defaults might prompt one to wonder how much default bias varies across domains.
Similar research, and experimental research, suggests that any number of things might make a difference, including the type of decision or subject matter of the default; the framing (the way the default and the alternatives are presented); the current default itself; users’ beliefs about the riskiness of the current default (or of switching); the source of the default; and the familiarity of the user with the type of choice at-issue. Much of this is a way of saying that context matters, which also includes the notion that “the value of an option depends not only on the option in question, but also on the other options in the choice set.”
Regarding Rangel’s expert opinion, and given the many factors that might influence a default’s “stickiness,” one might wonder whether Rangel ever studied search or other pre-loaded phone or desktop defaults directly. And I suppose he might have, although what we know of his testimony doesn’t describe any such research, or cite it, and I cannot find any good candidates on his Google Scholar publications page (a go-to for me, but just one, and I don’t think I have a default for such things). I’m guessing it’s a “no,” but I don’t know that for certain.
Google’s default deals have been likened to slotting fees paid for desirable grocery-shelf placement (see Brian Albrecth here, and with Thom Lambert here), which he has studied, if not in actual grocery-store settings. As I discussed last November, he’s conducted (with co-authors), various:
studies of mechanisms of visual attention, with particular experiments displaying certain food choices to investigate—e.g., the influence of visual salience on the amount of attention devoted to a display. Generalizing findings from, e.g., cereal-box placement to the durability of search-engine defaults seems a stretch (or entirely speculative).
For example, experiments on visual saliency and consumer choice (e.g., here, here, and here) suggest that the visual properties of, e.g., packaging, can influence visual attention, which can bias choice in experiments where items were displayed in marked screen “shelves” (not actual grocery shelves), and eye tracking was measured. Rangel and colleagues observed that eliminating peripheral display of alternatives approximately doubled the attention biases—the amount of attention devoted to a displayed food item. The authors “suggest that individuals might be influenceable by settings in which only one item is shown at a time, such as e-commerce.”
Perhaps, but it’s not really a measurement of the extent to which, say, varied cereal displays in grocery stores (for which manufacturers might pay slotting fees) actually modulate the choices made by either marginal consumers or those with established preferences. Manipulating the relative amount of attention that research subjects pay to appetitive (desirable) and aversive food items, it was observed that appetitive items were 6-11% more likely to be chosen with long fixation (exposure)—with a similar but somewhat different bias for aversive items. What that says about the degree to which a given default setting for a browser or search engine might or might not be a barrier to switching is anyone’s guess. All of it is very context-dependent. For example, visual salience has an especially significant effect when consumers are (experimentally) forced to make very rapid choices. That’s interesting, and intuitive, but it constrains generalization across both experimental and real-world contexts.
In an artificial experimental setup presenting completely different choices, 6% to 11% seems less spectacular than the numbers on German organ-donation choices. And if we ask how rapid, we note that one study has participants exposed to images of cereal boxes on a screen for exposure times ranging from 70-500 milliseconds. That is, the very longest exposure was half a second.
I’m not suggesting that the research is unserious or that it has no use—not at all. Such studies do contribute to our overall picture of decisionmaking and choice in various ways, and building such a picture is a painstaking, complex, and piecemeal process.
But I do find myself wondering about its application. I don’t doubt that defaults matter, but I’m wondering how much. And if all available research suggests that the answer is “that depends,” (and that the most striking results are obtained with rushed or peripheral exposures, or in novel or perceived high-risk situations), then it seems to me that the expert testimony may have overgeneralized (or, to put it another way, grossly exaggerated how much it knows about the specific case at-issue, and how well it knows it).
Overgeneralization in search of explanatory power seems to be one of those things that happens periodically in the behavioral and brain sciences, from behaviorism, to the “new look” psychology, “visceral learning,” prospect theory, etc. I wrote about it in one specific context here, but that’s a very general caution, not a proof of anything in particular.
What difference does it make? Quite a bit, potentially, at least under a “but for” showing of causation, and perhaps not only under that standard. While we’re at it, what is the “but for” counterfactual? Are we wondering what would have happened if (a) there were no default general search engine for a given browser, (b) there were a different default general search engine, (c) the terms of the contract, and nature of the default, were different (and if so, in what regards), or (d) the path to switching defaults were different (perhaps easier, but again, in what ways, and how much easier)? Or something else?
We know that consumers can switch, even on mobile devices (and recall Rangel’s testimony that “Search engine default effects have stronger effects on mobile devices than on personal computers.”). As I (and many others) have noted, Android devices in Europe are required to offer a choice screen: no default. The consumer benefits of the mandate are unclear. Google’s share of general search on European mobile phones is reported to be about 96%. And while Microsoft preloads its Edge browser and Bing search engine as defaults on computers with the Windows operating system, Bing still lags well behind Google on general search on U.S. desktops. Both observations suggest that it’s possible—at least possible—for consumers, and not just tech-savvy consumers, to switch to whatever search engine they prefer.
Mehta didn’t just find that Google acquired market share by innovating; he found that they have continued to innovate. Moreover, evidence at trial suggested that Apple and others thought so too. The section of Mehta’s opinion on product development begins thusly: “Google is widely recognized as the best GSE available in the United States.” For this proposition, he cites, among other things, testimony from Apple, Mozilla, Motorola, T-Mobile, and AT&T.
And recall that monopolization is “distinguished from growth or development as a consequence of a superior product, business acumen, or historic accident.” There was ample evidence of product superiority (certainly, in terms of consumer preference) and business acumen, as well as missteps by competitors, such as Microsoft.
As Geoff Manne notes:
Judge Mehta holds that Google’s distribution deals had anticompetitive effects “because they ensure that half of all [general search engine] users in the United States will receive Google as the preloaded default on all Apple and Android devices.” He derives this conclusion (which he repeats several times) from the testimony of one of the plaintiffs’ economic experts, Michael Whinston, who finds that “50% of all queries in the United States are run through the default search access points covered by the challenged distribution agreements.” (internal citations omitted).
But that cannot be right. People can and do switch from defaults. Under the right conditions, most people will switch from a default (recall the share of U.S. desktop searchers run through Google on Windows machines). Add to that Judge Mehta’s findings that most U.S. consumers prefer Google, and it’s quite a leap to suppose that Google’s competitors are foreclosed from every single consumer running every single search on an Apple or Android machine. I know this, in part, because I’m typing on an Apple laptop on which I have–mirabile dictu–searched via Bing, DuckDuckGo, and any number of specialized search engines. And I have marched through the steps of changing defaults, just for giggles (just for giggles ex ante – ex post, it all went just fine, but wasn’t especially funny).
That’s anecdotal evidence, to be sure, but it’s mirrored by what’s known about the search behavior of hundreds of millions of consumers. The real question gets back to causal effects (and how sticky is it?), and what difference it would make. Given consumer preferences, that’s unclear.
If the but-for world were that of a U.S. version of the EU choice screen, would that make a difference? What sort of difference? If a very small percentage of users at the margin were to switch (say, to the highest option on the choice screen), would that make a significant competitive difference? Would it help the average consumer? Many consumers prefer a default—a highly functioning and simple package out-of-the-box. What if the but-for world were one in which no firm could bid for default status, but device manufacturers could pre-load a default (without receiving payment for it)? Given evidence introduced at trial, it’s not much of a stretch to imagine that Apple would still choose to preload Google.
One sort of potential difference might go to the question of minimum efficient scale. That is, could some intervention by the court enable competitors to reach minimum efficient scale, at which they might compete, even if they did not, in the near term, capture the leading market share? That’s part of the case that the DOJ brought, but on that, their complaint reads like a lot of hand waving over a difficult and fundamentally technical matter.
I led off with a joke about Bing, but Bing has made some interesting innovations employing LLMs, and these might matter going forward. Bing has captured some market share, even if many users of Microsoft’s Edge browser end up searching via Google. OpenAI is testing SearchGPT and others will no doubt enter the fray. But new technologies raise the question what the minimum efficient scale might be for new applications. Microsoft is a behemoth because it has developed and marketed useful products: it has hundreds of millions of users—huge troves of user data and the capital to acquire more. Again, we’d like to know what scale we’re talking about, and what difference the conduct made.
In brief, it’s easy to imagine that the default agreements made some difference (and not just to Apple’s bottom line), but the question of which difference, and what the difference would mean to competition and consumers, seems fundamentally up in the air. Under Microsoft-plus-Rambus, as I (and Geoff and others) read them, those facts should matter, and the burden was on the DOJ to make a stronger theoretical and empirical case. Perhaps even under the gummy—I mean, edentulous—standard, which is not simply one of establishing that the conduct could have made some difference.
All of that goes to the question of liability, but it also points to the next big question: What’s the remedy? The court hasn’t yet said. Tim Wu opines in The New York Times in favor of two radical remedies. He suggests, first, that the court might “force Google to divest from its web browser, Chrome, and the Android operating system for mobile devices.” In the alternative, it might “force Google, which has provided access to some of its A.I. language models, to grant anyone free and open access to all of its A.I. technologies, as well as the vast troves of data on which those technologies are trained.”
Right. The court might hand all of Google’s intellectual property to that plucky little startup from Redmond, Washington called Microsoft. The proposed remedies don’t target the default agreements at the heart of the dispute, and it’s hard to imagine that such interventions would leave competition and consumers better off. Part of the problem with the case, from the start, was the question what sort of remedy would, supposing one was called for.
But Wu’s column advocates for more than a remedy. As he sees it, the task at-hand involves both the restoration of competition and “to punish Google for its past offenses.” Namely, its default-placement contracts with Mozilla, Apple, and other device manufacturers. Which, allegedly, made it unduly hard for people to decide to try Bing and consider whether they might prefer it to Google. Which, as Judge Mehta notes, they don’t. Then again, to be fair, sometimes it’s hard to tell when Tim is joking.
On the other hand, the Times also reports that “[t]he Justice Department and state attorneys general are discussing various scenarios to remedy Google’s dominance in online search, including a breakup of the company.” And while they might not seek it—and may or may not want it. They shouldn’t want it–we shouldn’t want it–but I don’t expect that they’re joking.