Countless digital goods are free. Companies such as Facebook and Twitter obtain revenues from other sources, such as advertising. But in light of recent controversies, there have been practical discussions about changing the business model and more theoretical discussions about economic valuation. What if people were required to pay to use Facebook? How much would they be willing to spend?
Any answers would tell us something important about the value of social media in general. They might also tell us something more general about economic valuation of legal entitlements, about the potentially expressive quality of some consumption decisions, and about the disparity between traditional economic measures and actual human welfare. They would bear on potential legal regulation as well.
In April 2018, I conducted a pilot experiment to obtain some preliminary answers. Using Amazon’s Mechanical Turk, I asked 439 Facebook users—not a nationally representative sample, but one with significant demographic diversity—to say how much use of the platform is worth. More specifically, I asked 215 Facebook users a simple question: “Suppose that you had to pay for the use of Facebook. How much would you be willing to pay, at most, per month?” At the same time, I asked 234 other Facebook users a different question: “Suppose that you are being offered money to stop using Facebook. How much would you have to be paid per month, at a minimum, to make it worth your while to stop using Facebook?”
It will be noticed that the first question asks about willingness to pay, whereas the second asks about willingness to accept. According to standard economic theory, the two questions should produce the same answers. But behavioral economists have shown that in important contexts, they do not. In many experiments, willingness to accept is about twice as much as willingness to pay – a reflection of the endowment effect, which means that the allocation of the initial entitlement affects people’s valuations. In brief: people tend to place greater value on things they already own. For example, many people would pay less to buy a lottery ticket than they would demand to give up a lottery ticket that they own. Contractual default rules also give rise to an endowment effect. One question is whether an endowment effect would be observed for digital goods; another question is its magnitude.
For the first survey question, asking about the maximum amount that people would be willing to pay to use Facebook, the median answer was just $1 per month. The average was $7.38. Most strikingly, nearly half of participants (46 percent) said that they would pay no more than $0 for a month of Facebook use! (Recall that they were all Facebook users.)
For the second question, asking about the minimum amount that people would accept to stop using Facebook, the median answer was $59 per month. The average was $74.99. It should be clear that the disparity between willingness to pay and willingness to accept is unusually large. (The numbers were roughly the same for Democrats and Republicans, and also for men and women.) We might describe this disparity as a “superendowment effect,” greatly exceeding the one-to-two endowment effect ratio often observed in previous studies, and in contrast to the “no endowment effect” observed for money tokens, for goods held for resale, and sometimes, for goods with well-established economic values. (For illuminating discussions about the underpinnings and domain of the endowment effect, see here, here, and here.)
By way of comparison, consider the environmental setting, where unusually large disparities have also been observed between willingness to pay and willingness to accept. One study found that people would demand about five times as much to allow destruction of trees in a park as they would pay to prevent the destruction of those same trees. When hunters were questioned about the potential destruction of a duck habitat, they said that they would be willing to pay an average of $247 to prevent the loss—but would demand no less than $1044 to accept it. In another study, participants required payments to accept degradation of visibility ranging from 5 to more than 16 times higher than their valuations based on how much they were willing to pay to prevent the same degradation.
I will return shortly to the environmental domain. In my study, the most obvious puzzle is the very low median for willingness to pay. For Facebook users, could continued Facebook use really be worth no more than a single dollar per month? It is plausible to think that for many digital goods (including other social media platforms), a similarly low willingness to pay would be observed, at least in surveys. We might speculate that the reason is expressive, and so not at all a reliable measure of the welfare benefits of using Facebook. Having had to pay nothing for Facebook, people dislike the idea of a monthly fee. When almost half said that they are willing to pay $0, they were most likely giving a protest answer, announcing: “If you are going to start charging me, well, then, forget about it!”
Something similar might be said about those who said that they would pay only a small monthly amount (say, $10). They might well have been saying that they resent having to pay for what they previously received for free. Here, then, is a reason to think that the low median willingness to pay does not offer adequate information about the welfare effects of using Facebook.
Return to the environmental studies in this light. It would be easy to imagine studies of clean air or clean water that would also generate puzzlingly low willingness-to-pay figures, and for the same reasons: A good once enjoyed for free is now being subject to some kind of charge. But in the environmental studies listed above, the real puzzles come from the high willingness to accept numbers. In general, such numbers can be a questionable proxy for welfare effects. When people say that they would demand a very high amount of money to give up some good that they own (coffee mugs, lottery tickets), they might not be focused on what they could do with that money. In the environmental context, there is also a moral dimension that could inflate willingness to accept responses; respondents might not want to be responsible for the destruction of environmental amenities, whatever the effects of those amenities on respondents’ welfare.
By contrast, willingness to pay is more likely to activate consideration of opportunity costs. For that reason, there is reason to doubt whether the $59 median, in response to the second question, is sufficiently informative about the welfare benefits of using Facebook.
There is also an independent point to consider. For my second question, some or perhaps many participants might have resented the very idea that “someone” is trying to pay them to stop using the platform. Their resentment might well have manifested itself in a high figure.
These points suggest severe limitations to both willingness to pay and willingness to accept surveys as measures of the welfare effects of digital goods that have formerly been provided for free. Expressive answers might well be provided for willingness to pay questions, and resentment might infect answers to willingness to accept questions. In real markets, of course, different results might be expected. Some media outlets, such as the New York Times and the Washington Post, have shifted to require paid subscriptions, rather than providing free content (as they previously did). In surveys, elicited willingness to pay might have been far lower than actual willingness to pay as observed through behavior. For subscribers to formerly free services, initial resentment, resulting in some kind of expressive reaction, might recede in favor of a welfare calculation, in which people decide how much the good is worth to them. It remains to be determined when and by how much willingness to pay or willingness to accept figures, elicited in surveys, would differ from those that are observed in behavior.
In a much more elaborate study, Brynjolfsson et al. tried to value use of Facebook by asking consumers if they would prefer (a) to maintain access to the platform or (2) to give it up for one month in response to a specified payment. Brynjolfsson et al. also used a large, nationally representative sample. With their method – a “discrete choice experiment” – they asked people to choose between two identified options and to specify the one they valued more. It is important to see that a discrete choice experiment ought to avoid some of the distortions of both willingness to pay and willingness to accept. The reason is that it asks for a direct comparison between options and identified amounts of money (as opposed to someone, such as Facebook, asking for payment, or by someone, such as an outsider, specifically asking to pay). At the same time, discrete choice experiments cannot avoid an endowment effect: The relevant questions will be asked to people who are, or are not, current “owners” of the good at issue.
Brynjolfsson et al. used a large, nationally representative sample, limited to Facebook users. The median answer was in the vicinity of $40 to $50 to give up Facebook for a month (slightly below the median answer to my second question).
Aware of various technical limitations in their study, Brynjolfsson et al. do not insist on those particular numbers. But they do urge that digital goods, including social media, are producing large, monetizable benefits that are not included in conventional measures of well-being, such as gross domestic product. That conclusion is both important and plausible. Nonetheless, it is important to add two qualifications.
The first, signaled by my pilot study, is that whatever numbers are generated will be an artifact of the particular method that is used. If different methods produce different numbers, then it will be challenging to decide which one is the best measure of economic value. For goods that have been provided free, willingness to pay numbers might not be reliable, because they might well reflect resentment about being asked to pay for such goods. Willingness to accept numbers are better, but they have the standard problems (including opportunity cost neglect). If the goal is to capture welfare effects, discrete choice experiments are probably best, but insofar as the relevant questions are posed to current users, they will embody a kind of endowment effect.
The more fundamental question is that we need better measures of the effects of such goods on people’s experienced well-being. Brynjolfsson et al. title their impressive paper, “Using Massive Online Choice Experiments to Measure Changes in Well-Being,” but well-being is emphatically not what they are measuring. At best, they are measuring predictions of well-being.
People might be willing to pay $10 each month for the right to use Facebook, or demand $60 to give up that right. In discrete choice experiments, the median value might be $50. But what are the effects of Facebook on their actual experience? Are they enjoying life more, or less, or the same? Those are the more important questions. Willingness to pay and willingness to accept numbers, and the outcomes of discrete choice experiments, are best understood as reflecting people’s predictions about effects on well-being, translated into monetary terms. (And in analyzing uses of social media, it is also important to ask whether, for some users, the relevant platform is in the nature of an addiction, and not increasing welfare at all.) The actual effects are the gold standard; they are what matter.
Information about those effects is starting to emerge (see, e.g., here, here, and here). The results are both complicated and mixed. Use of Facebook and other social media platforms may well have different effects on different demographic groups. It certainly has different effects on different components of well-being. Different uses of Facebook, and different ways of spending time on the platform, undoubtedly have different effects on users’ well-being. For both private and public institutions, there is a pressing need to learn more.
This essay should be taken as a preliminary statement about issues and problems that I hope to explore in far more detail. I am grateful to Daniel Kahneman and Richard Thaler for valuable discussions; neither of them should be held responsible for my mistakes and confusions. Thanks to Arevik Avedian for indispensable help with the pilot survey described here. The Behavioral Economics and Public Policy Program at Harvard Law School provided support.
Over the last year, I have served as a consultant to Facebook on several occasions, but not in connection with any of the issues and experiments discussed here.