Progress Snapshot 5.10 from The Progress & Freedom Foundation

A recent telephone poll conducted by professors at Berkeley and the University of Pennsylvania concluded, “Contrary to what many marketers claim, most adult Americans (66%) do not want marketers to tailor advertisements to their interest.” The study’s authors claim that their poll is the “the first nationally representative telephone (wireline and cell phone) survey to explore Americans’ opinions about behavioral targeting by marketers.” They also assert that the poll indicates that “if Americans could vote on behavioral targeting today, they would shut it down.” Advocates of regulating online data collection have trumpeted this poll as evidence consumers demand legislation to protect their privacy. “This research gives the F.T.C. and Congress a political green light to go ahead and enact effective, but reasonable, rules and policies,” declared Jeff Chester, a leading critic of online advertising.

But what is most surprising about this poll is not that 66% of users said they do not want tailored online ads, but that 34% of users said they did! The key, initial question of “whether or not you want the websites you visit to show you ads that are tailored to your interests,” presents no trade-off. The fact that anyusers said “yes” indicates that many users paused to do the rough mental math about the unarticulated trade-off between the benefits of receiving tailored ads and the costs of that tailoring.

The methodology of opinion polls necessarily affects respondents’ mental calculations, rendering polls not just easily manipulated, but inherently unreliable as indicators of real preferences. Every poll reflects the bias of its authors to some degree by the way questions are worded, the order in which they are asked, the sample surveyed, etc. The easiest way to bias the results of a poll is to omit any mention of the trade-offs at issue. This poll simply buried the issue of trade-offs in a heavily loaded follow-up question: After telling respondents that marketers “often use technologies to follow the websites you visit and the content you look at in order to better customize ads,” the interviewer asked whether the respondent would allow advertisers to “follow [them] online in an anonymous way in exchange for free content.” Only 10% of users said they would allow this voluntary exchange.

What does this tell us about whether, and how, government should further regulate online advertising? Precious little: Not only does this poll overstate the costs of targeted advertising, understate its benefits, and ignore the tools available to users to address their privacy concerns but, like any opinion poll, this one tells us more about the psychology of decision-making under the artificial uncertainty of polls than about the choices users would actually make in the real world.

User Uncertainty About Concepts Like “Tailoring” and “Following”

Even the word “tailoring”—though benign compared to other words the study’s authors could have used ( e.g., “track,” “monitor,” “record”)—is so vague as to leave respondents wondering what it really entails. One can only speculate as to what users thought the word meant (since the poll did not ask), but it seems likely that some of these scarier words probably flashed through the minds of respondents in the instant before they answered the question. Indeed, the word “tailoring” conflates both the costs and benefits of personalized advertising in a single, vague word. Given this ambiguity, it’s hardly surprising that most users would say “no”—not just to receiving tailored advertising (66%), but also to receiving tailored discounts (49%) and news (57%). If users had been asked about receiving “relevant” (rather than “tailored”) ads, the responses probably would have turned out somewhat differently—just as an additional 17% of users agreed to receiving tailored “discounts,” whose value to users is more readily apparent: saving money on potential purchases.

The second set of questions asked users whether it “Would be OK… if these ads [discounts/news] were tailored for you based on following what you do on the website you are visiting… [24% said yes] OTHER websites you have visited… [34% said yes] and OFFLINE—for example, in stores? [25% said yes].” Again, the term “follow” was not defined. A third set of questions explained to respondents that marketers “often use technologies to follow the websites you visit and the content you look at in order to better customize ads.” The interviewer then asked whether the respondent would “definitely allow, probably allow, probably NOT allow, or definitely not allow advertisers” to “follow you online in an anonymous way in exchange for free content”—and only 10% of users said yes. Thus, it appears that users are more, not less, hostile to tailored advertising when reminded of the trade-offs involved (35% yes in the first set of questions, 10% yes in the third). What explains this paradox?

The most obvious explanation is that, by the time the respondent got to the critical question about “allowing” tailored advertising, they had heard the word “follow” at least five times: once in each of the three questions about whether tailoring was OK, once in the introduction about how marketers customize ads and once in the question itself—each time increasing uncertainty as to how “tailoring” really works and more than negating any suggestion of “anonymity.” Furthermore, asking users whether something should be “allowed” implies that there are undisclosed reasons why it should not be. This much is simple psychology—obvious to anyone who wanted to craft a poll that would support a particular regulatory agenda.

But behavioral economics research tells us something even more profound about the way our brains work: human beings hate making choices, and loathe uncertainty even more. Indeed, such “mental accounting” or “mental transaction” costs appear to be the primary reason why, after a decade of efforts to develop a micropayments system that can fund online content and services, no such system has emerged—and thus why Internet publishers instead rely primarily on advertising revenues ($23.5 billion in 2008) to fund “free” offerings for consumers. In this case, merely forcing consumers to consider the costs of “tailoring” and being “followed,” and decide whether these things are “OK” or should even be “allowed” strongly tips the scales in favor of the outcome desired by the study’s authors because these considerations and decisions are significant psychological costs in themselves, which likely outweigh the diffuse benefits of tailored advertising, which users simply do not appreciate.

Indeed, the scale tips so strongly that the study suggests that 73% of Americans object to having ads tailored based on “what you do on the website you are visiting.” Would not this objection apply to purely contextual advertising “tailored” to the keywords entered by a user in a search engine or to the keywords that appear on a particular page to which a user has navigated within a site? If so, this study isn’t just about the bogeyman of “behavioral” advertising, but about essentially all online advertising, which is to some degree “tailored.” Indeed, must lawmakers protect us from the tailoring of news (71%) and discounts (62%) within websites? Or, if data collection is the real harm to consumers, what about the fact that hundreds of millions of people happily share far more personal information every day on social networks or using grocery discount cards? Opinion polls simply cannot answer these questions.

The Direct Benefit of Tailored Ads: Relevance

Whatever Americans tell pollsters about “tailored” ads, they also complain about irrelevant ads: A previous poll found that 72% of consumers “find online advertising intrusive and annoying when the products and services being advertised are not relevant to [their] wants and needs” and 85% say that less than 25% of the ads they see while browsing online are relevant to their wants and needs. Real-world experiments confirm that users reveal a clear preference for more relevant advertising. In a 2004 experiment, click-through rates (CTR) for behaviorally targeted ads were between 94% and 225% higher than for contextually targeted ads. A 2009 study found that the difference could be between 670% and 1000% percent, depending on how well-tailored the ads were. In other words, users in the real world were two to eleven times more likely to click on highly-tailored ads. Truly, actions speak louder than words: Users clearly “vote with their clicks” for ads they find relevant—i.e., they vote for “tailoring.”

Further reinforcing this conclusion is the fact that better tailoring increases not only click-through rates but also “conversion rates”—the percentage of users who actually complete the action desired by the advertiser, whether that be making a purchase or signing up for a list. A 2008 experiment found increased conversion rates of 400-900% (2008). This indicates that relevant ads really do help consumers find things they like—and that they like the fruits of tailoring, however they respond when asked about “tailoring” as an abstract concept that conflates costs (“How are they following me?”) and benefits (“What’s in it for me?”).

The Indirect Benefit of Tailored Ads: Free Content & Services

Even less apparent to poll respondents than the direct benefit of tailoring (increased relevance) are the indirect benefits: In particular, greater relevance to the user means more effective communication for the advertiser, and increased ad revenue for most online publishers per ad on their sites. Thus, there exists a clear quid pro quo: in effect, users “pay” for content and services by sharing information about their interests. Even more fundamentally, users “pay” for content by seeing ads. But both quid pro quos are implicit: Users can simply choose not to “pay” by using readily available tools in their browser to blocking ads and/or tracking. In essence, today’s system allows users who don’t like ads—tailored or otherwise—to opt out at little or no cost, much as if they simply decided not to pay for a product they bought at their local grocery store.

This creates a serious dilemma, given that advertising increasingly stands alone as the lifeblood of online content and services. Indeed, ads have long funded the costs of generating content for radio, television, and newspapers (with subscriptions paying only for distribution). The basic reason is simple economics: In competitive markets, prices tend to fall to the marginal cost of production. The Internet has simply borne this theory out in full:

1. Producing the first unit of content (e.g., a news story or video) remains costly, so while the marginal cost of every additional unit is essentially zero,average cost is not.
2. The failure of micropayments online seems to confirm that, no matter how low the technological transaction costs are, the mental transaction costs involved combined with even tiny payments will exceed the perceived value of most content.
3. The world of media scarcity in which consumers could choose from only a few sources of content (e.g., news, entertainment) has given way to a world of staggering media abundance and the choices of users are no longer constrained by the tyranny of physical limitations like distance and printing costs.
4. Because pure information cannot be copyrighted (and fair use allows significant referencing and quotation), very little content is so unique that users cannot find a ready substitute elsewhere if a site (or even cartel of sites) attempted to charge.

These forces have given birth to the world of “Free,” where few (if any) users will pay for something they can get for nothing. While there are a number of ways to fund content and services, advertising is far and away the leading business model for the new economy: Indeed, overall advertising market is expected nearly to double its share of total U.S. ad spending from 8.7% in 2008 ($23.4 billion) to 15.2% ($37.2 billion). But with 44% of advertising revenue going to search engines (which show highly “tailored” ads simply based on search terms), hundreds of thousands of publishers—from the mightiest to the tiniest—rely on $7.6 billion (33% of the total) in “display” ad revenue. Yet this base is tiny: Most websites earn a fraction of the revenue generated by offline ads: roughly $0.60 to $1.10 per thousand impressions (CPM) online versus average CPMs of $4.54 (radio) to $10.25 (broadcast). This unprofitability of online advertising, and the fact that certain kinds of online content (e.g., video and online services) does not provide the textual keywords necessary for basic contextual targeting is driving publishers to ad networks that offer behavioral targeting, which is expected to grow from $525 million in 2007 to $4.4 billion in 2012—when it will represent 25% of all display ad spending.

In short, advertising is indispensable to the future of online media, but it is also currently inadequate to sustain “Free” culture. As Adam Thierer and I warnedearlier this year: “The advocates of regulation pay lip service to the importance of advertising in funding online content and services but don’t seem to understand that this quid pro quo is a fragile one: Tipping the balance, even slightly, could have major consequences for continued online creativity and innovation… Something must give because there is no free lunch.” In 2001, long before Google mattered and before he worked for them, Kent Walker (now Google’s general counsel) put it best in a seminal law review article:

Privacy is both an individual and a social good. Still, the no-free-lunch principle holds true. Legislating privacy comes at a cost: more notices and forms, higher prices, fewer free services, less convenience, and, often, less security. More broadly, if less tangibly, laws regulating privacy chill the creation of beneficial collective goods and erode social values… Such regulation would likely increase both direct and indirect costs to the individual consumer, reduce consumer choice, and inhibit the growing trend of personalization and tailoring of goods and services.

Thus, as Jim Harper and Solveig Singleton concluded in their 2001 paper With a Grain of Salt: What Privacy Surveys Don’t Tell Us:

privacy surveys in particular… suffer from the “talk is cheap” problem. It costs a consumer nothing to express a desire for federal law to protect privacy. But if such law became a reality, it will cost the economy as a whole, and consumers in particular, significant amounts that surveys do not and cannot reveal.

We Need a Behavioral Economics Experiment, Not Just Another Poll

The Berkeley-Penn poll could certainly have done more to present these trade-offs to respondents and less to color their responses by inflating mental transaction costs. But even the most “fair” poll cannot meaningfully simulate the trade-offs inherent in the real world. If we really want to know how muchsubjective value consumers place on a particular aspect of their privacy, we must look to the preferences they reveal in the process of making real choices.

Of course, the best experiment is the one being conducted in the real world every day. No laboratory experiment can ever fully replicate all of the conditions of the real world, but a behavioral economics experiment could tell us more about the revealed preferences of Internet users than any poll. Unlike the real world, an economist could vary certain conditions in a lab experiment to tell us how various changes to current industry practice, user empowerment, or user education might actually affect real consumer choices. At a minimum, any experiment would require the following to inform policymaking about online advertising and privacy.

First, the experiment should vary the mechanisms by which notice is provided to users as to how tailoring works ( e.g., placement, interface, wording) and what those notices actually say.

Second, test subjects must make real choices in real use of the Internet with trade-offs in real money and their own time between either paying for access to a particular site or getting access for free in exchange for receiving tailored ads based on at least the three variables presented as questions in the Berkeley-Penn study: (i) users’ browsing activity on that site; (ii) their browsing activity on other sites; and (iii) offline activity or demographic information.

The second variable is critical because it addresses the value created by behaviorally tailored ads, which could be wiped out by regulation. Search engines are able to sell highly effective advertising based solely on information provided directly to the site (search keywords, which are highly indicative of user interest), and some sites can sell lucrative advertising based on purely contextual targeting because their content contains keywords that advertisers value highly ( e.g., a site for digital camera enthusiasts). But the vast majority of websites, and especially non-commercial websites, would produce little ad revenue if advertisers could only guess at the likely interests of visitors based on the keywords on that site. This, in a nutshell, is why so many sites stand to gain so much from behavioral targeting—particularly in the Internet’s “Long Tail.” To be useful, an experiment must reflect this dynamic.

In the real world, of course, it might be possible for the user to opt-out of tracking without losing access to content because today’s quid pro quo is implicit and most sites operate on a “No Cost Opt-Out” basis for tracking and even seeing ads. But in order to tell us how much consumers really care about tracking, the experiment must place some value on access to content that is supported by free content and services.

Third, the experiment must examine the extent to which user empowerment affects user choice: If some users are uncomfortable with having their browsing activity tracked, is it because they are concerned about all tracking or only tracking of certain sensitive activities, such as researching medical issues or—everyone’s favorite—viewing pornography? How does the availability of privacy management tools change user choices about ad-tailoring? Do Americans really want tailoring banned, or do they just want the ability to exercise easy choice about when they want to participate? How would those choices change when they come at a cost (e.g., seeing more ads) and privacy-sensitive users cannot simply free-ride off the value created by users whodon’t opt-out of targeted advertising (and also don’t block ads)?

Such an experiment would, by its very nature, be imperfect—but far less imperfect than any poll about opinions on privacy. Until a proper experiment is conducted by trained behavioral economists, all we can say with confidence is the following:

1. Users don’t understand exactly how ads are tailored;
2. Users seem to be concerned about “tailoring” or “following” in the abstract;
3. Users are generally unwilling to pay for online content and services; and
4. Better tailoring of ads means more funding for content and services.

There is only one approach that can address all these concerns: educate users about how online advertising works and how they can implement their own privacy preferences, while constantly striving to further empower users to make privacy management easier.

The Obama administration has been greeted with enthusiasm by scientists who see the potential for “research-based policy.” Reason, not ideology, will govern. The New Scientist, among other zines, headlines “Let Science Rule: the Rational Way to Run Societies.” (May 28, p. 40-43) This is part of a larger theme: Behavioral economics is taking off.

One commonly offered example of policy fixes that are crying out for a research-based approach is sex education. Abstinence-only sex education programs are well-taxpayer–funded at the federal, state, and international level. And they don’t work, either for HIV prevention or pregnancy prevention. What advocates of abstinence perhaps forgot is that the social context in which abstinence was preached with some success to upper-middle-class Victorian young ladies (not the young men, even the Victorians had more sense than that) were perpetually accompanied by adult chaperones. (The result was horrendous… the innocent young ladies would ultimately be infected with venereal diseases by their husbands, and they and their babies would suffer and often die without ever being told what was wrong or how it could have been prevented–bringing us to an important chapter in U.S. free speech history, as the “birth controllers” and other advocates like Katharine Hepburn’s mother fought for an end to the silence). Done. Criminal law also could reap substantial benefits from a research-based approach. I have written elsewhere about the problems of ignoring deterrence research in copyright.

But it gets harder. The key problem: There is research, and then there is research. Much of it is done by advocates or just people who are careless with their assumptions. Some of these people might not even be aware of the extent to which they are advocates.

Examples: One is the EU FLOSS Report, which I was reading the other day. Some interesting data, but the authors are so busy making the case for open source they neglect key questions. Such as, just for example, how do you measure the contribution to GDP of volunteer labor? Can one simultaneously express concern (as the FLOSS report does) about the EU’s low levels of investment in software, and delight that open source reduces the amount that firms must spend in software research or on software? If one wishes to make the cheery prediction that increasing the take-up of open source software within Europe will close their innovation gap with the United States, shouldn’t one consider that perhaps the United States might simultaneously increase it’s take-up rate if it appears to be a good idea? Just why does the United States seem to be more innovative than Europe anyway?

And so on. On the problem of result-oriented studies, New Scientist quotes Laurence Moore of Cardiff University in the United Kingdom: “They’re almost designed to show that the idea is a good idea … Rigorous evaluations are perceived as threatening rather than supportive of better policy.”

Sound studies that run contrary to popular ideas are often simply ignored. Abstinence policy is one example. But the problem is not confined to conservatives. Another example is Head Start. This is the classic 1960’s early childhood research-based triumph. Except the only study showing lasting results from Head Start was an study designed by the originator of the program. Later studies continue to cite the original study, and to cite studies citing the original study, and so on. The results showing long term gains have never been replicated. This problem, too, is recognized: “Assessing social policies using randomised controlled trials did start to take off in the US from the 1960s to 1980s. But the practice has declined, partly because policy makers became disenchanted when the trials did not endorse their brainwaves, according to Sheila Bird, a statistician at the Medical Research Council Biostatistics Unit in Cambridge, UK.”

This raises the question of why politicians and many other policymakers stick with such determination to their agendas even in the face of contrary evidence. Another way of asking the same thing: What is ideology and what does it do for us? I offer some thoughts. Are people just dumb? Is it ego? People must “save face” and are unwilling to back down from a position once taken publicly?

Some of our willingness to go with general principles is a good thing. A vast amount of human experience gets summed up and expressed in the form of ideology. The United States Constitution is an ideological document. Yet it is also based on human experience with hundreds of years of monarchy, condensed into few words. One does not, and ought not, lightly set such things aside. Example: There is a not-well-enough-known Supreme Court case, Buchanan v. Worley. At issue was the constitutionality of racial segregation laws. The supporters of the laws were filed many research reports from highly progressive social scientists, purporting to prove that segregation was good for people. Quite rightly, the Supreme Court dismissed the research and favored the principle of equality in the eyes of the law. Segregation statutes were unconstitutional. Without the Court’s willingness to declare them so, the United States could have developed a full-blown apartheid system along the lines of South Africa.

Certainly it would be good to scrutinize the human experiences that go into our ideologies and rules of thumb carefully. But this cannot always be done. I do not wish to be a gulag guinea pig.

Also, there are general, as well as particular lessons to be drawn from research-based policy. One general lesson appears to be that many bright ideas fail. The excellent book Seeing Like A State elaborates on this theme. But there is another. Why is it the need for research-based policy so pressing and not, say, just for example, research-based ideas for small business or how to cook a good hard-boiled egg? Why is the need usually in the public sector, not the private sector? When research is needed in the private sector, such as medicine—why is it taken up readily, egos set aside, while the public sector has been so stubborn?

This is not merely an accident. In the private sector, failure often has natural and severe consequence for those who support or act on a bad idea. In government, failure often has no consequences except embarrassment for those who act on or support a bad idea. Research will continue to be ignored without accountability. Which brings us back to ideology; a gentle rule of thumb favoring action through the private sector, not the public sector, may be more research-based than some would like to think after all.

My original post is here at convergence law.com.

The rural broadband debate has been in the news a lot lately. Yesterday, DSL Reports ran a story sharply criticizing a report released by the US Internet Industry Association (an ISP lobbyist firm). But as Ars pointed out, the report actually offers some facts revealing that broadband availability in the U.S. isn’t nearly as bad some have suggested.

79 % of homes with a phone line can now get DSL, and 96 % of homes with cable can get broadband. Considering just about every home has a phone line, and most people have cable, these numbers suggest the main reason for the lack of rural broadband users isn’t the lack of availability, but the lack of adoption. Of course, rural areas have slower speeds and higher prices than urban areas. This makes sense, because building out a network in low-density areas costs more per subscriber versus urban areas, where a single apartment complex can house hundreds of users.

Still, groups argue that massive government subsidies are needed to promote broadband deployment in rural areas. ConnectedNation (a Washington-based non-profit) released a report a couple weeks ago, “The Economic Impact of Stimulating Broadband Nationally”, which concluded that accelerating broadband could pump $134 billion into the U.S. economy.

According to the report, high-speed web access is massively beneficial to subscribers. The report, which obtained most of its data from surveys of broadband subscribers, found that the average broadband user reports savings of $217 a year “as a direct result of becoming healthier through obtaining healthcare information online.” On top of this, the average user reports driving 102 fewer miles per month on account of shopping online.

These figures strike me as quite unrealistic. No surprises there—economists find that when responding to surveys, consumers notoriously overestimate their true reservation price. To be sure, web access can reduce the frequency of trips to the doctor or the store. But annual savings of $217 in medical costs, and 102 fewer miles driven per month, are inconceivable to me—and I visit WebMD and shop online all the time.

If consumers value broadband so much, why are so few of them willing to pay the premium for living in rural areas? You’d think that if signing up for broadband lets people cut back on mileage, save time, and reduce medical bills, that even at $75 or $100 a month broadband should be a real bargain for consumers living in the countryside. Yet many rural residents still decide not to get broadband.

In reality, subscribers probably vastly overestimate the value of high-speed Internet access. Actions speak louder than words, and actual broadband statistics suggest that people living in the countryside value broadband a lot more when somebody else is paying for it.