It is very easy to posit a sense of exploitation when it comes to data. Try this for size: “Networks are relying on data by individuals but they aren’t being paid and this is making those networks profitable.” Substitute “plantations” for “networks,” “labour” for “data,” and “individuals” for “slaves” and you get the picture.
Of course, you might object that there is a big difference here in that people who use networks today have a choice as to whether to do so. But if those services are necessary for operating even at a modest level in a modern economy (as they arguably are for banking if not for social media), then the notion of choice is not that powerful when simple optimisation makes it look like you really have no choice. In other words, Janis Joplin had it right: you aren’t that free if you have lots to lose. That’s an idea that is at the core of economic thinking. Mark Zuckerberg famously remarked that he didn’t “feel” like he ran a monopoly but similarly many of his customers don’t “feel” like they have much choice.
Will paying people for data help?
The proposed answer to this concern is that people should be paid for the data they provide. To be sure, those putting this idea forward — for instance, Eric Posner and Glen Weyl in their new book, Radical Markets — understand that people are ‘paid’ to provide data because they receive services — often free services — in return. But they are big believers in markets. When there are no prices one of the consequences is that you do not get an efficient allocation of effort and resources. But they go further and see that the redistribution of rents that would come from paying people for data might resolve or mitigate coming disruptions to labour markets from the influx of artificial intelligence and automation.
As I see it, the problem with these proposals is that they do not work through the pricing effects of them. I think they may be subtle. Let me explain by using a very simple model.
Suppose that a service has value v(x) – c.x to a consumer who provides an amount (or set) of data, x. The cost to them of providing x is c ‘per unit’ of data where c relates to inconvenience or concerns for privacy/security. For the moment, let’s suppose this is a paid service (I’ll come back to an advertising-based ‘free’ service below). And just to make it fun, let’s assume that there is a single provider of the service so it is a monopoly.
What can that monopolist do? The monopolist can compell a consumer to pay a price, p, for the service and to provide data at level, x, if they want to access the service. In this case, the maximum price the monopolist can charge is p = v(x) – c.x and if they want to maximize profits they will want to compell an amount of data so that v(x) – c.x is maximized. Let’s call this x* in which case, the monopolist earns profits of v(x*) – c.x*.
Now suppose that we require the monopolist to pay for data at a regulated rate of t per unit of x. In this case, the monopolist’s profits are p – t.x but the consumer receives v(x) – c.x + t.x – p. As the monopolist sets price, it can now charge p = v(x) – c.x + t.x so its profits will be v(x) – c.x and it has an incentive to still set x at x*. The monopolist earns profits of v(x*) – c.x* and consumers earn nothing. In other words, the whole system is neutral in t. This is a common result in economics and arises all of the time. But the implication is that requiring firms to pay consumers for data will change nothing. This is because that consumer is ultimately the consumer of data as well while the firm just provides the means of making that happen.
What if we took away the monopolist’s ability to compell provision of data? In other words, we could leave it up to the consumer as to how much data to provide. It is pretty easy to see that this won’t change anything. In this situation, if a consumer chooses to pay p for the service, they will choose to provide data to maximize, v(x) – c.x – p which gives x* as before. Knowing this, the firm will set p = v(x*) – c.x*. Thus, the very same outcomes will arise. (Notice: that this assumes the firm sets its price before the consumer chooses how much data to supply).
What about both a regulated payment and consumer control of data provision? In this case, if a consumer chooses to pay p for the service, they will provide data to maximize v(x) – c.x + t.x – p. In this case, the consumer will choose to provide more data to the firm as the payment for it offsets some and perhaps more than their costs of providing it. Call the new optimal provision x** (> x* if t is positive). In this case, the firm will charge p = v(x**) – c.x** + t.x** for the service. Their profit will be v(x**) – c.x** (which is lower than before) while the consumer still receives zero. In other words, t is no longer neutral but the effects of having it are all bad for efficiency without any impact on distribution. Put simply, you want someone to feel the true costs of data provision and the payment to consumers actually subverts that.
What about for advertising-based services?
All the action in the above argument occurred because the consumer was paying an explicit price for the service. What happens when that is not the case? For instance, suppose that the monopolist chooses p = 0 but earns money through advertising (or other data selling) which itself is improved if data is provided. Suppose that this revenue is r(x).
What happens in this case? Let me summarize the different outcomes:
- Monopolist chooses x: the monopolist will want more x to increase r(x) but the consumer will choose not to be part of the service if v(x) > c.x. Thus, the maximum x the monopolist can compell is x’ where v(x’) = c.x’. This gives the monopolist r(x’) and the consumer 0.
- Monopolist chooses x and t: the monopolist may actually want to pay the consumer for more data. In particular, by choosing t, the monopolist can select the level of data it wants. That is because v(x) = (c – t)x, there exists a t that can give the monopolist any amount of data for a price of t = c – v(x)/x. The monopolist, therefore, earns r(x) – t.x = r(x) + (v(x) – c.x). It, thus, chooses x to maximize joint surplus. The consumer still ends up with zero. The reason the monopolist may not do this is because it already obtains the data that would maximize profits without setting t > 0. That is, the value of its service already implies that consumers receive sufficient value themselves from providing more data.
- Monopolist must pay t for x: in this case, the monopolist would like to choose x so that r'(x”) = t. If at this point, v(x”) > (c – t)x” then this is feasible. The monopolist earns r(x”) – t.x” and the consumer earns v(x”) – (c – t)x”. However, it is also possible that, for the consumer, v(x”) < (c – t)x” in which case a lower x will be provided, the consumer will receive 0. In effect, a regulated t will only change things if it is above the level the monopolist would set themselves. In that case, it will reduce the amount of data employed in the service. That is, it acts in a similar way to a minimum wage.
In this particular case, it is not feasible to have a situation where consumers receive a regulated payment for data and can choose the level of data they provide. This is because they will opt to supply more data than the firm wants at that price. In the previous examples, with paid services, there was a check on the data a consumer chose to provide because it would impact on what they paid for the service. That constraint does not exist here.
What this analysis implies is that the case for paying people for data is equivalent to the case for a minimum wage. It can provide them with more surplus but it is likely to reduce the amount of data firms choose to use. However, this case only arises in situations where consumers are not themselves paying for the services but firms are making money elsewhere.
Not so fast!
There is something incomplete about the above argument. The data people are providing services is often personal data — in particular, the reason why r(x) is increasing in x is because advertising becomes more effective when it can target individuals directly. But that value that advertisers are paying for is not coming out of thin air. In fact, it is coming because of interactions with those individuals themselves.
Why does an advertiser pay r(x) to a network for access to that particular consumer? The straightforward reason is that r(x) represents the additional profits the advertiser may get from the consumer. There are many ways we can view this. But let’s assume that an advertiser will create value by matching with a consumer and that better ads can facilitate that match. Suppose the probability of a match is M(x) and if it is made the consumer gets a share, a, of the product value, V, while the advertiser gets 1 – a. The advertiser’s product if they pay for an ad, M(x)(1-a)V – r(x). Obviously, if the network is a monopolist they can set r(x) = M(x)(1-a)V.
The earlier analysis shows that if t can be chosen by the platform, the amount of data will be that which maximizes joint surplus of v(x) – c.x + M(x)V; characterised by v'(x) + M'(x)V = c. If the consumer portion (v(x) – c.x) is positive, then the network can achieve this even if t = 0. On the other hand, if t is regulated, then the monopolist will choose x so that M'(x)(1-a)V = t. The consumer will earn v(x) – c.x + M(x)aV + M'(x)x(1-a)V. Thus, the consumer benefits from the regulation to the extent they are paid a share of the advertising revenue but area harmed to the extent that the network purchases or uses a smaller amount of data.
In summary …
Herein lies the insight: to the extent that having consumers not be paid for their data is an indication that they value the use of their data by the network, by forcing the network to pay for that data, the consumer can be made worse off because they can no longer just give the data to the network. Thus, the whole analogy with slavery or the supply of pure labour breaks down because the consumer may want to encourage the network to make use of more of its data. Requiring the network to pay subverts that process.
In order for the notion of regulating payments for data makes sense, you have to believe that consumers do not gain utility by giving additional data to networks. Some (and perhaps many) consumers do give their data freely to networks now. Thus, it is entirely possible that they will be made worse off it networks are required to pay them for that data because those networks may structure themselves to no longer make use of the data rather than pay for it.
That said, there is more to be done here. I have sketched out some arguments that show that a simple approach to considering payments for data does not appear to be the correct approach. There may, however, be solid rationales that could justify it. At the moment, however, I would like to see that grounded in a model and supported with evidence before proposing such payments on what appears to be “moral” grounds.