More than a hundred million US households have fixed-line broadband access, generating tens of billions of dollars in yearly access charges, but there is surprisingly little reliable knowledge about online user behavior. Nonetheless, copious amounts of attention — not to mention subsidies and other government resources — are applied to Universal Service policies and other regulatory and judicial activities (such as merger and antitrust evaluations) that must hinge on a detailed understanding of how consumers respond to various internet service providers’ offers. Effective decisions require policymakers to have a detailed knowledge of how consumers allocate their attention across a panoply of internet applications and content.
How consumers allocate their attention between internet sites may seem to have much in common with other standard consumer choice settings: Internet user attention is a scarce resource and users must make choices about where to spend their limited time. Yet consumer choice across internet sites lacks one of the standard hallmarks of a market: relative prices reflecting scarcity and directing the allocation of a scarce resource. Most households pay a fixed fee for monthly service and then allocate online time among endless options without further expenditure. Unless a household faces a binding cap on usage, no price shapes any other marginal decision. Instead, choice depends on non-monetary considerations and the gains of the next best choice. Evidence suggests that, until recently, only a small fraction of users faced monetary constraints when deciding whether to use additional online resources. Only one of the top twenty websites currently (Netflix) is a subscription service where the access price plays an explicit role in decision-making.
Probing consumers’ online behavior
In this context, a recent paper by Andre Boik, Shane Greenstein, and Jeffrey Prince sheds a great deal of light on what has to date been a rather dark corner for broadband regulatory policy and antitrust decision-making. Using extensive microdata from a service that tracks household-level demographics and internet usage data over an entire year, the authors compare the internet usage choices on the “home internet device” (that is, fixed-line broadband usage) made in any week, by over forty thousand households in 2008 and over thirty thousand in 2013.
The paper uses two traditional metrics — the total amount of time spent online and the market shares of that time by site category — and two novel measures assessing the breadth and depth of user attention. Concentration of attention across sites (breadth) is measured using a Herfindahl-Hirschman Index where the market share of each site visited is equal to the total number of minutes spent at that site divided by the total minutes spent at all sites. The average expenditure per site visit (depth) is measured by the fraction of site visits that exceed ten minutes.
This time it’s no different?
Consistent with earlier work by Goldfarb and Prince, the authors find that in both 2008 and 2013 total internet attention (hours spent online in an average week) decreases as household income increases (Figure 1). Importantly, major shifts in attention have occurred between application categories, as users have moved away from chat and news sites towards social media and video (Figure 2). However, the researchers observed no meaningful changes in either the breadth or depth of attention (Table 1). These key measures of “how” households allocate attention online are roughly constant across time.
Figure 1: Total time online by income (2008, 2013)
Figure 2: Changes in share of attention across the top 1000 sites by category (2008, 2013)
That is, the range of sites visited (breadth) and the fraction of site visits exceeding ten minutes (depth) on fixed-line household access devices have stayed remarkably similar, despite huge changes in the sites actually visited and increasing internet use on mobile devices between the two years when user behavior was assessed (2008 and 2013). Furthermore, the measures are remarkably constant across different types of household (measured by a range of demographic characteristics including age and education of the household head, household size, and the presence or absence of children). That so little has changed despite the addition of extensive mobile device usage is especially remarkable.
Table 1: Summary statistics for attention measures
Significant implications for policy and industry
Universal Service policies for telephony services presumed that reducing prices to the underserved would increase uptake, but that the usage of these marginal consumers will be less costly to service than that of the average consumer because they could not afford to make as many calls. The households typically targeted by such policies were those with lower incomes, older residents, or more people (particularly children). These same assumptions may not hold when addressing universal service for broadband. Using income as a targeting characteristic may lead to more consumers with higher-than-average service demands (and expected service costs) joining. Furthermore, the limited connection between age and income with expenditure per site visit suggests little (usage) cost difference for service to the elderly, drawing into question the merits of traditional policies of varying prices for the underserved by age.
With regard to competition policy, typical modelling using cross-price elasticities is not relevant in a world of online attention where the price of additional access is zero. Competition may be driven as much by substitution across sites that fit an available time slot as by product characteristics. This makes the minimum attention required to consume the site as well as switching costs measured in time more relevant as metrics when evaluating mergers and antitrust cases.
With regard to advertising, the study confirms what online advertising brokers have long known — that household demographics are of little use when scheduling advertising. Knowing which sites attract low breadth and high depth visitors may be more useful when trying to minimize the number of duplicate exposures, as much as knowledge about the particular depth and breadth characteristics of a specific user can influence ad targeting.
Time to start measuring something different
In sum, this paper suggests some merit in regulatory authorities such as the Federal Communications Commission collecting information on internet usage factors, including those required to calculate the authors’ novel breadth and depth measures, to assist policy, antitrust, and business decision-making.