Digital Discrimination: The Case of Airbnb.com

Edelman, Benjamin, and Michael Luca. “Digital Discrimination: The Case of Airbnb.com.” Harvard Business School Working Paper, No. 14-054, January 2014.

Online marketplaces often contain information not only about products, but also about the people selling the products. In an effort to facilitate trust, many platforms encourage sellers to provide personal profiles and even to post pictures of themselves. However, these features may also facilitate discrimination based on sellers’ race, gender, age, or other aspects of appearance. In this paper, we test for racial discrimination against landlords in the online rental marketplace Airbnb.com. Using a new data set combining pictures of all New York City landlords on Airbnb with their rental prices and information about quality of the rentals, we show that non-black hosts charge approximately 12% more than black hosts for the equivalent rental. These effects are robust when controlling for all information visible in the Airbnb marketplace. These findings highlight the prevalence of discrimination in online marketplaces, suggesting an important unintended consequence of a seemingly-routine mechanism for building trust.

Comments to European Commission on Proposed Google Commitments with Zhenyu Lai

The European Commission last month posted a restatement of its concerns at certain Google practices as well as Google’s proposed commitments. This week I filed two comments, including one with Zhenyu Lai, critiquing Google’s proposal. They are available here:

Comments on AT.39740 (Edelman and Lai) as to Google’s exclusive use of screen space to promote its own specialized services, and as to an alternative remedy to preserve competition and user choice in the area of specialized search services.

Comments on AT.39740 (Edelman) as to the failure of Google’s proposed commitments to undo the harm of Google’s past violations, and alternative remedies preserve competition in the area of specialized search services, taking data from publishers, providing advertising services to publishers, and allowing advertisers to use multiple ad platforms.

Our suggested remedy for Google favoring its own vertical search services: Let users choose from competing offeringsAs to Google’s decision to favor its own specialized search services, Zhenyu and I question whether Google’s proposed commitments would effectively preserve competition. In our recent measurement of the effects of Google Flight Search, we found that Google’s large Flight Search "onebox" displays sharply reduced traffic to other online travel agencies as well as driving up the proportion of clicks to AdWords advertising. Google proposes to display a few small "Rival Links" to competitors, but these links would be placed and formatted in ways that make them unlikely to attract users’ attention or facilitate competitive markets.

Instead, we suggest a "ballot box" in which users can choose their preferred specialized search services in any area where Google offers and favors its own specialized search services. For example, the first time a user runs a hotel search or flight search, the user would be presented with a menu of options — reminiscent of the "browser ballot box" a user sees when first booting Windows in Europe. This approach is entirely feasible: Facebook has long merged content from myriad independent developers; numerous developers quickly built browser add-ons to disable Google Search Plus Your World when they found its initial results unhelpful; and the G++ for Google Plus browser add-on integrates other social networks with G+ even though Google declines to provide such integration. Indeed, search engine guru Danny Sullivan in September 2011 suggested that Google "let[] people choose their shopping, local, etc one box provider?", echoing my February 2011 call for interchangeable components for competing specialized search services. In our comment, Zhenyu and I explore a proposed implementation of this concept.

As to the Commission’s concerns more generally, my comment addresses Google’s failure to undo the harm resulting from Google’s past violations. For each of the Commission’s concerns, I present alternative remedies focused on protecting competition and affirmatively undoing the harm from Google’s past violations. I also flag the need for tough penalties to deter further violations by Google and others.

Reuters yesterday reported that the Commission is likely to require further concessions from Google, including improved remedies. Comments to the Commission are due by June 27.

Google’s Exclusive Flight Search OneBox with Zhenyu Lai

Google often shows “OneBox” search results promoting its own services. These results have prompted antitrust scrutiny: Google awards these preferred placements exclusively to Google’s own services, such as Google Flight Search and Google Maps, but never to competing services such as Kayak or Mapquest. Moreover, Google presents OneBox with special format, including distinctive layouts, extra images, and even in-page interactivity — benefits not available to ordinary listings for other sites. Regulators and competitors sense that these exclusive practices can undermine competition and innovation by denying traffic to would-be competitors. But how large is the effect? How much does Google’s exclusive OneBox placement impact search engine traffic to adjacent online markets?

In a working paper, Zhenyu Lai and I measure the impact of OneBox by using a quasi-experiment before and after the introduction of Google Flight Search. Using a third-party data service, we compare user behavior on searches across thousands of search queries like “cheap flights from sfo to san ” (which displayed a OneBox for Google Flight Search), and similar search queries like “cheaper flights from sfo to san” (emphasis added) (which did not display OneBox). We find that Google’s display of Flight Search in an exclusive OneBox decreased user click-through rates on unpaid search results by 65 percent, and increased user click-through rates on paid advertising links by 85 percent. This effect was disproportionately evident among online travel agencies that were popular destinations for affected search queries.

Our draft provides detailed empirical results as well as a model of how a search engine’s incentives to divert search depend on consumers’ perceptions of the difference between non-paid and paid placements.

Exclusive Preferential Placement as Search Diversion: Evidence from Flight Search

(update: published as Edelman, Benjamin, and Zhenyu Lai. “Design of Search Engine Services: Channel Interdependence in Search Engine Results.” Journal of Marketing Research (JMR) 53, no. 6 (December 2016): 881-900.)

Revisiting Search Bias at Google

Last week Joshua Wright posted a critique of my January 2011 Measuring Bias in ‘Organic’ Web Search (with Ben Lockwood). Some quick thoughts –

First, there’s some important common ground here: Wright and I both find that Google shows many of its own results, and does so in prominent positions.

Now, Wright says Bing presents its own results slightly often more than Google does so. In contrast, important portions of my analysis indicated that own-service links are particularly prominent at Google. Why the gap between my analysis and Wright’s? One key challenge is the lack of a natural basis of comparison. Suppose Google controlled 50% market share while five competitors held 10% each. Then we could compare Google’s results to a possible consensus among the others — better revealing whether and when Google favors its own services. But in fact the runners-up are much smaller: after Bing+Yahoo, we reach smaller firms like Blekko, each with market share far below 1%. Without a competitive marketplace providing a baseline, comparisons between search engines are necessarily difficult. So it’s no surprise that the numbers come out differently depending on the approach.

Wright criticizes my decision to examine brief, popular searches rather than a selection of actual user searches. But searches are messy and idiosyncratic: Each day at Google, 20% to 25% of searches are completely new, never before seen. Wright favor analysis of searches seen in AOL logs, but this method tends to emphasize unusual searches like "dog who urinate on everything" and "you’re pregnant he doesn’t want the baby" (the first two examples in Paul Boutin’s 2006 examination of AOL logs). In contrast, I chose to focus on short, simple searches where biased results have a particularly broad effect.

Wright argues that if Bing presents its own results as often as Google does, then own-service links must be pro-competitive, raising no antitrust concern. I disagree. The same behavior can have very different consequences when performed by a dominant firm versus a smaller competitor. Indeed, section 2 of the Sherman Act only applies to companies with market power. Meanwhile, companies without market power may engage in the exact conduct that Section 2 prohibits. For example, Microsoft faced antitrust litigation when it included Internet Explorer in Windows, even as Apple permissibly included its Safari browser in MacOS.

Wright suggests that antitrust investigation of Google is stillborn for lack of consumer harm. I see two problems with this argument. First, competitive foreclosure is a sufficient cause for concern. Certainly Google’s own-service links can stymie competition: It’s a tall order to start the next Yelp if Google may adjust its algorithm to always put Google Places first. From 2006 through June 2011, ads from Google Checkout merchants featured a special logo — a benefit unavailable to merchants using competing checkout systems (like PayPal). Even the perception of such favoritism can disrupt competition: In June 2011 a Google Offers salesperson told a merchant that signing up with Offers would provide "SEO benefits" to make the merchant "number one in Google." Google quickly disavowed that statement, but on Wright’s theory, this tactic would be entirely permissible. Imagine the harm to Groupon, LivingSocial, and fellow travelers if Google gave Offers advertisers the favored map placement that AdWords advertisers already enjoy. In my view, that’s the wrong result. New providers necessarily rely on Google to reach users, but their business plans won’t work if Google can systematically favor its own services at competitors’ expense.

Second, when Wright says there’s no consumer harm, he has the wrong consumers in mind. The folks paying the bills for Google are advertisers — and advertisers pay a high price that has only grown as Google gains market share.

Ultimately, Google’s antitrust problems go far beyond algorithmic search preference. Google’s harsh treatment of advertisers smacks of market power; an advertiser with a real choice of ad networks would not accept Google’s high prices and one-sided terms. Google’s dealings with mobile handset makers similarly draw on Google’s dominance: If there were numerous popular third-party operating systems for mobile handsets, Google wouldn’t be able to compel manufacturers into dropping third-party software like Skyhook; and if there were numerous strong vendors for search, maps, videos, and other core mobile apps, Google wouldn’t be able to bundle its mobile apps to compel handset manufacturers to take all of these as a condition of preinstalling any of them. At every turn, we see Google leveraging its dominance in certain sectors to shore up its position in others – and that’s a n approach that rightly raises significant antitrust concern.

Remedies for Search Bias

Disclosure: I serve as a consultant to various companies that compete with Google. But I write on my own — not at the suggestion or request of any client, without approval or payment from any client.

In a forthcoming paper (update, November 2011: paper is available), I’ll survey the problem of search bias — search engines granting preferred placement and/or terms to their own links or to others’ links chosen for improper purposes. What purposes are improper? Given others’ work in that area, I’ll defer my thoughts on that subject to the paper. Today I’d like to focus on remedies — what tactics a dominant search engine ought not employ due to their detrimental effects on competition, and how prohibiting those tactics would help assure fair competition in search and related businesses.

The prospect of legal or regulatory oversight of search results has attracted skepticism. A search industry news site recently questioned the wisdom of investigating search bias by arguing that, even if bias were uncovered, “it’s not clear what any remedy would be.” James Grimmelmann last month critiqued the suggestion that search engines can be biased, and he argued that even if such bias exists, the legal system cannot usefully prevent it. Discomfort with the prospect of legal intervention extends even to those who ultimately see a need for oversight: For example, Pasquale and Bracha title a recent paper Federal Search Commission?, ending the title with a question mark to credit the immediate shortfalls of an overly bureaucratic approach. Meanwhile, Google’s caricature of regulation warns of government-mandated homogeneous results and unblockable web spam, offering a particularly pronounced view of search regulation as intrusive and undesirable.

I envision an alternative approach for policy intervention in this area — addressing the improprieties that various sites have alleged and stopping specific practices that ought not continue, while avoiding unnecessary restrictions on search engines’ activities.

Experience from Airline Reservation Systems: Avoiding Improper Ranking Factors

A first insight comes from recognizing that regulators have already — successfully! — addressed the problem of bias in information services. One key area of intervention was customer reservation systems (CRS’s), the computer networks that let travel agents see flight availability and pricing for various major airlines. Three decades ago, when CRS’s were largely owned by the various airlines, some airlines favored their own flights. For example, when a travel agent searched for flights through Apollo, a CRS then owned by United Airlines, United flights would come up first — even if other carriers offered lower prices or nonstop service. The Department of Justice intervened, culminating in rules prohibiting any CRS owned by an airline from ordering listings “us[ing] any factors directly or indirectly relating to carrier identity” (14 CFR 255.4). Certainly one could argue that these rules were an undue intrusion: A travel agent was always free to find a different CRS, and further additional searches could have uncovered alternative flights. Yet most travel agents hesitated to switch CRS’s, and extra searches would be both time-consuming and error-prone. Prohibiting biased listings was the better approach.

The same principle applies in the context of web search. On this theory, Google ought not rank results by any metric that distinctively favors Google. I credit that web search considers myriad web sites — far more than the number of airlines, flights, or fares. And I credit that web search considers more attributes of each web page — not just airfare price, transit time, and number of stops. But these differences only grant a search engine more room to innovate. These differences don’t change the underlying reasoning, so compelling in the CRS context, that a system provider must not design its rules to systematically put itself first.

I credit that some metrics might incidentally favor Google even as they are, on their face, neutral. But periodic oversight by a special master (or similar arbiter) could accept allegations of such metrics; both in the US and in Europe, a similar approach oversaw disputes as to what documentation Microsoft made available to those wishing to interoperate with Microsoft software.

Evaluating Manual Ranking Adjustments through Compulsory Disclosures

An alternative approach to avoiding improper ranking factors would require disclosure of all manual adjustments to search results. Whenever Google adjusts individual results, rather than selecting results through algorithmic rules of general applicability, the fact of that adjustment would be reported to a special master or similar authority, along with the affected site, duration, reason, and specific person authorizing the change. The special master would review these notifications and, where warranted, seek further information from relevant staff as well as from affected sites.

Why the concern at ad hoc ranking adjustments? Manual modifications are a particularly clear area for abuse — a natural way for Google to penalize a competitor or critic. Discourage such penalties by increasing their complexity and difficulty for Google, and Google’s use of such penalties would decrease.

I credit that Google would respond to the proposed disclosure requirement by reducing the frequency of manual adjustments. But that’s exactly the point: Results that do not flow from an algorithmic rule of general applicability are, by hypothesis, ad hoc. Where Google elects to use such methods, its market power demands outside review.

Grimmelmann argues that these ad hoc result adjustments are a “distraction.” But if Google’s manual adjustments ultimately prove to be nothing more than penalties to spammers, then regulators will naturally turn their attention elsewhere. Meanwhile, by forcing Google to impose penalties through general algorithms rather than quick manual adjustments, Google will face increased burdens in establishing such penalties — more code required and, crucially, greater likelihood of an email or meeting agenda revealing Google’s genuine intent.

Experience from Browser Choice: Swapping “Integrated” Components

Many complaints about search bias arise when longstanding innovative services are, or appear to be at risk of becoming, subsumed into Google’s own offerings. No ordinary algorithmic link to Mapquest can compete with an oversized multicolor miniature Google Maps display appearing inline within search results. (And, as Consumer Watchdog documented, Mapquest’s traffic dropped sharply when Google deployed inline maps.)

On one hand it is troubling to see established firms disappear in the face of a seemingly-insurmountable Google advantage. The concern is all the greater when Google’s advantage comes not from intrinsic product quality but from bundling and defaults. After all, if Google can use search to push users to its Maps product, Maps will gain market share even if competitors’ services are, on their merits, superior.

Yet it would be untenable to ask Google to disavow new businesses. It is hard to imagine a modern search engine without maps, news, or local search (among other functions largely absent from core search a decade ago). If legal intervention prevented Google from entering these fields, users might lose the useful functions that stem from integration between seemingly-disparate services.

What remedy could offer a fair chance of multiple surviving vendors (with attendant benefits to consumers), while still letting Google offer new vertical search services when it so chooses? E.C. antitrust litigation against Microsoft is squarely on point, requiring Microsoft to display a large choice screen that prompts users to pick a web browser. An initial listing presents the five market-leading options, while seven more are available if a user scrolls. But there is no default; a user must affirmatively choose one of the various options.

Taking the “browser choice” concept to search results, each vertical search service could, in principle, come from a different vendor. If a user prefers that her Google algorithmic search present embedded maps from Mapquest along with local search from Yelp and video search from Hulu, the user could configure browser preferences accordingly. Furthermore, a user could make such choices on a just-in-time basis. (A possible prompt: “We noticed you’re looking for a map, and there are five vendors to choose from. Please choose a logo below.”) Later, an unobtrusive drop-down could allow adjustments. The technical barriers are reasonable: External objects could be integrated through client-side JavaScript — just as so many sites already embed AdSense ads, YouTube player, and other widgets. Or Google and contributors might prefer server-to-server communications of the sort Google uses in its partnerships with AOL and with Yahoo Japan. Either way, technology need not stand in the way.

I credit that many users may be content with most Google services. For example, Google Maps enjoyed instant success through its early offering of draggable maps. But in some areas, Google’s offerings have little traction. Google’s Places service aspires to assess quality of restaurants and local businesses — but Yelp and Angie’s List draw on specialized algorithms, deeper data, and longstanding expertise. So too for TripAdvisor as to hotel reviews, and myriad other sites in their respective sectors. A user might well prefer to get information in these areas from the respective specialized services, not from Google, were the user able to make that choice.

Google often argues that competition is one click away. But here too, the E.C.’s Microsoft litigation is on point. Users had ample ability to install other browsers if they so chose, but that general capability was not enough when the standard operating system made one choice a default. Furthermore, at least Windows let other browsers truly immerse themselves in the operating system — as the default viewer for .HTML files, the default application for hyperlinks in email messages, and so forth. But there is currently no analogue on Google — no way for a user, even one who seeks this function, to combine Google algorithmic search with a competitor’s maps, local results, or other specialized search services.

Banning Other Bad Behaviors: Tying

Using its market power over search, Google sometimes pushes sites to adopt technologies or services Google chooses. Sometimes, Google’s favored implementations may be competitively neutral — simply technical standards Google wants sites to adopt (for example, presenting an index of pages to Google’s crawlers in a particular format). But in other instances, Google uses its power in search to promote adoption of Google’s own services.

I first flagged this tactic as to Google Affiliate Network (GAN), Google’s affiliate marketing service. GAN competes in one of the few areas of Internet advertising where Google is not dominant, and to date Google has struggled to gain traction in this area. However, Google offers remarkable benefits to advertisers who agree to use GAN: GAN advertisers alone enjoy images in their AdWords advertisements on Google.com; their advertisements always appear in the top-right corner above all other right-side advertisements (never further down the page); they receive preferred payment terms (paying only if a user makes a purchase, not merely if a user clicks; paying nothing if a user returns merchandise, a credit card is declined, or a server malfunctions). Moreover, merchants tend to use only a single affiliate network; coordinating multiple networks entails additional complexity and risks paying duplicate commissions on a single purchase. So if Google can convince advertisers to use GAN, advertisers may well abandon competing affiliate platforms.

Google’s tying strategy portends a future where Google can force advertisers and sites to use almost any service Google envisions. Google could condition a top AdWords position not just on a high bid and a relevant listing, but on an advertiser agreeing to use Google Offers or Google Checkout. (Indeed, Checkout advertisers who also used AdWords initially received dramatic discounts on the bundle, and to this day Checkout advertisers enjoy a dramatic multicolor logo adjacent to their AdWords advertisements, a benefit unavailable to any other class of advertiser.) Google would get a major leg up in mobilizing whatever new services it envisions, but Google’s advantage would come at the expense of genuine innovation and competition.

Measuring Bias in “Organic” Web Search with Benjamin Lockwood

By comparing results between leading search engines, we identify patterns in their algorithmic search listings. We find that each search engine favors its own services in that each search engine links to its own services more often than other search engines do so. But some search engines promote their own services significantly more than others. We examine patterns in these differences, and we flag keywords where the problem is particularly widespread.

Even excluding “rich results” (whereby search engines feature their own images, videos, maps, etc.), we find that Google’s algorithmic search results link to Google’s own services more than three times as often as other search engines link to Google’s services.

For selected keywords, biased results advance search engines’ interests at users’ expense: We demonstrate that lower-ranked listings for other sites sometimes manage to obtain more clicks than Google and Yahoo’s own-site listings, even when Google and Yahoo put their own links first.

Details, including methodology, analysis, and policy implications:

Measuring Bias in “Organic” Web Search

Bias in Search Results?: Diagnosis and Response

Edelman, Benjamin. “Bias in Search Results?: Diagnosis and Response.” Indian Journal of Law and Technology 7 (2011): 16-32.

I explore allegations of search engine bias, including understanding a search engine’s incentives to bias results, identifying possible forms of bias, and evaluating methods of verifying whether bias in fact occurs. I then consider possible legal and policy responses, and I assess search engines’ likely defenses. I conclude that regulatory intervention is justified in light of the importance of search engines in referring users to all manner of other sites, and in light of striking market concentration among search engines.

Hard-Coding Bias in Google “Algorithmic” Search Results

I present categories of searches for which available evidence indicates Google has “hard-coded” its own links to appear at the top of algorithmic search results, and I offer a methodology for detecting certain kinds of tampering by comparing Google results for similar searches. I compare Google’s hard-coded results with Google’s public statements and promises, including a dozen denials but at least one admission. I conclude by analyzing the impact of Google’s tampering on users and competition, and by proposing principles to block Google’s bias.

Details, including screenshots, methodology, proposed regulatory response, and analogues in other industries:

Hard-Coding Bias in Google “Algorithmic” Search Results

Competition among Sponsored Search Services

Edelman, Benjamin. “Competition among Sponsored Search Services.” U.S. House of Representatives, Committee on the Judiciary, Task Force on Competition Policy and Antitrust Laws, 2008. (Hearing cancelled.) (Reprinted in Working Knowledge: Google-Yahoo Ad Deal is Bad for Online Advertising.)

Last month I was asked to testify to the United States House of Representatives Committee on the Judiciary Task Force on Competition Policy and Antitrust Laws about competition among paid search providers, particularly the proposed Google-Yahoo partnership.

At the last minute, the hearing was cancelled, and I won’t be able to testify at the rescheduled session. Rather than let my draft written statement languish, I’m taking this opportunity to post the prepared testimony I had planned to offer:

Competition among Sponsored Search Services.