Distinguishing bandwidth and latency in households’ willingness-to-pay for broadband internet speed

https://doi.org/10.1016/j.infoecopol.2018.07.001Get rights and content

Highlights

  • First paper to measure relative willingness-to-pay for broadband internet latency, upload and download bandwidth, and data caps.

  • Willingness-to-pay for bandwidth and data caps is highly concave, with the exception of added value for unlimited data.

  • The FCC's Phase II auction for broadband provision subsidies penalizes latency much more than consumer willingness-to-pay would suggest.

  • Excluding latency from a willingness-to-pay survey for broadband can result in overvaluation of its speed counterpart, bandwidth.

  • Willingness-to-pay surveys similar to the one in this paper can help guide policy and private investment concerning characteristics of products and services.

Abstract

We measure households’ willingness-to-pay for changes in key home broadband Internet connection features using data from two nationally administered, discrete choice surveys. Both surveys include price, data caps, and download and upload bandwidth, but only one includes latency. Together, these surveys allow us to measure tradeoffs between bandwidth and other connectivity features such as price and data caps, and perhaps most notably, provide the only empirical evidence to date of tradeoffs between bandwidth and latency. We find that households’ valuation of bandwidth is highly concave, with relatively little added value beyond 100 Mbps. For example, households are willing to pay about $2.34 per Mbps ($14 total) monthly to increase bandwidth from 4 Mbps to 10 Mbps, $1.57 per Mbps ($24) to increase from 10 to 25 Mbps, and only $0.02 per Mbps ($19) for an increase from 100 Mbps to 1000 Mbps. We also find households willing to pay about $8.66 per month to reduce latency from levels obtained with satellite Internet service to levels more common to wired service. Household valuation of increased data caps is also concave as caps increase from 300 GB to 1000 GB, although consumers place a significant premium on unlimited service. Our findings provide the first relative valuation of bandwidth and latency and suggest that current U.S. policy may be over-penalizing latency relative to reductions in bandwidth and data caps. For example, we find that in its CAF Phase II Auction, the FCC is imposing a bidding penalty for latency that is about five times higher than what our WTP estimates suggest it should be relative to bandwidth offered.

Introduction

Broadband Internet is typically considered an always-on, high-speed Internet connection. Speed in particular has come to be viewed as especially important, with Internet Service Providers (ISPs) competing based on the speeds they can offer. As of 2015, ISPs were spending tens of billions of dollars annually on broadband investment, much of this going toward speed improvements (USTelecom). In addition, policymakers have become increasingly concerned about whether consumers are getting broadband that is “fast enough.” These concerns have led the Federal Communications Commission (FCC) to increase its definition of “broadband” (assessed in terms of bandwidth) several times—from 200 Kbps1 in either direction (i.e., downstream or upstream) in 1999 to 4 Mbps downstream and 1 Mbps upstream in 2010, to 25 Mbps down and 3 Mbps up in 2015.

FCC changes to the definition of broadband are not meaningless, bureaucratic exercises. They can, for example, affect how the Commission views competition, which affects merger and other decisions. In addition, the FCC allocates several billions of dollars per year through the Universal Service Fund, and the FCC must define the attributes of the services it will subsidize. Ideally, subsidies would correlate with how much consumers value the different features of broadband service.

Despite its prominence in policy and private investment, little is known about how much consumers value incremental improvement in bandwidth (particularly at the high end), and almost nothing is known about how they value other features of broadband speed and quality, such as latency and data caps. While bandwidth is the amount of data that can be transmitted over a connection in a given amount of time, latency is the time it takes for a data packet to make the round trip between the user's computer and another computer, typically a server located somewhere else. Thus, speed as understood colloquially is not just bandwidth, but a combination of bandwidth and latency. It is necessary to understand how consumers value those components separately in order to learn how they value “speed.” Furthermore, a clear understanding how consumers value the key components of Internet connection quality can provide important guidance for policy.

Our study is especially relevant, for example, to an upcoming reverse auction for universal service funds in unserved areas. In this auction, which we describe in detail below, bids will include not just the (hopefully) minimum amount a provider is willing to accept in subsidies in order to provide service, but also the bandwidth, latency, and data caps of the service the provider will offer. All of this information will be combined into a score that will determine the winning bids. Our study can help determine whether the relative weights on the various features of the bids correspond to consumers’ relative weights.

In this paper, we measure households’ valuations of Internet connection characteristics, focusing on bandwidth, latency, and data caps. To do this, we design and execute discrete choice surveys administered to more than 1,000 individuals. These surveys allow for variation in the key features of home Internet connection: download and upload bandwidth, latency, data cap, and price. A survey allows us to generate a degree of variation in bandwidth and latency that we generally do not see in the marketplace, thereby facilitating empirical analyses.2 We administer two surveys: one that includes latency as a connection feature and one that omits it (with 978 and 433 participants, respectively). By administering two separate versions of the survey, we can assess whether consumers value bandwidth differently depending on whether they also consider latency.

After collecting our survey responses, we estimate willingness-to-pay for home Internet features using standard discrete choice methods (a multinomial logit model). We find, similar to previous analyses, that consumers highly value bandwidth enhancements at lower levels, but the incremental value of bandwidth decreases rapidly. In particular, households on average value increasing bandwidth from 4 Mbps to 10 Mbps at about $14 ($2.34/Mbps), 10 to 25 Mbps at $24 ($1.57/Mbps), 25 to 50 Mbps at $14 ($0.57/Mbps), 50 to 75 Mbps at $8 ($0.32/Mbps), and 75 to 100 Mbps at $4 ($0.16/Mbps). Households were willing to pay only an additional $19 ($0.02/Mbps) for bandwidth increased from 100 Mbps to 1 Gbps.

Household valuation of allowed data transmission (data caps) is similarly concave, but consumers appear to place a premium on unlimited data transfer. Households valued an increase in a data cap from 300 GB to 600 GB at $12 ($0.04/GB), another $11 to increase the cap from 600 GB to 1000 GB ($0.03/GB), and an additional $35 for an increase from 1000 GB to unlimited. This premium on unlimited is especially noteworthy, since households use only 190 GB of data per month on average as of 2016 (Telecompetitor, 2016).

Meanwhile, households on average value a decrease in latency from 300–600+ ms (milliseconds) to under 10 ms at about $8 per month. Such a decrease in latency is approximately equivalent to moving from a satellite to a wired connection of the same bandwidth.

Our results across the two surveys (with and without latency) indicate that failing to account for latency increases measured Willingness-To-Pay (WTP) for download bandwidth, but has no effect on measured WTP for upload bandwidth or data cap. Hence, consumers may interpret download bandwidth as a proxy for the combined quality of download bandwidth and latency when latency is not expressed.

Our findings also allow us to compare household valuation of the features of broadband quality. For example, consumers value increasing bandwidth from 10 to 25 Mbps about three times more than they value reducing latency from satellite to wired levels. In other words, while households value latency improvements, they are generally viewed as notably less valuable than bandwidth and data cap improvements, at least across observed market levels (e.g., 1,000 Mbps download bandwidth, 100 Mbps upload bandwidth and unlimited data cap).

We also find sensible differences in valuation across subgroups divided by demographics and usage. For example, as expected, respondents engaged in tasks such as gaming and file transfers place greater value on speed than those who don't.

Our findings have several important implications. First, they provide valuable information for broadband market players (e.g., telcos, satellite, and cable companies) about the value – and ultimately the likely profits – of investments that generate high speeds, as well as which types of improvements generate the most value. In doing so, they provide the first relative valuation of bandwidth and latency. They also suggest that an omitted feature, at least in this setting, can lead to biased estimates, particularly when it is a component of a broader product attribute (speed). The bias appears to be contained to only a subset of the other features comprising “speed” (download bandwidth).

Importantly, our findings also help to inform policies aimed at generating higher available speeds by identifying the extent of at least some of the benefits, which can be weighed against the costs. They can be particularly informative for auction design when allocating broadband subsidies, where bids are handicapped based on speed and data-cap provision. Concerning the CAF Phase II auctions, we find that they implicitly valued latency improvements relative to bandwidth and data cap improvements at about five times what our valuation estimates indicate.

As noted above, the technology used to provide the broadband service is a major determinant of latency. Satellite service has substantially higher latency (> 300 ms) due to the need for signals to go to the satellite and back, compared to service using only terrestrial technologies (fiber, cable, copper, which are < 100 ms). Nonetheless, satellite could be a cost-effective solution to provide broadband services in rural or isolated areas, where deployment of other technologies can be expensive. Hence, there is a potential trade-off, where satellite technology can be a cheap solution to universalize the access to broadband Internet, but at the cost of a high latency. The FCC's auction handicaps this tradeoff, and our results help assess how the handicapping of latency relates to consumer valuation of latency.

Our paper relates to a substantial stream of literature on Internet demand, which has been measured in different ways, using both market data and survey/experimental data. Studies using market data include Goolsbee and Klenow (2006), who use the variation of opportunity cost, measured by individual forgone wages, to estimate Internet demand in terms of time use (hours). Nevo et al. (2016) use the variation of shadow price, observed from the usage-based broadband plan in multiple billing cycles, to estimate Internet demand in terms of data use. One of the two key speed factors, bandwidth, is partially captured in Nevo et al. (2016), which generates utility only through the reduction of the marginal cost on consuming more data. Ahlfeldt et al. (2016) use market data on housing prices to capture the part of consumer surplus generated via broadband access. Their results show that broadband access increases property price significantly, while the marginal effect diminishes to zero when the actual bandwidth reaches approximately 5 Mbps.3

Boik (2017) analyzes household cable and satellite broadband subscription data, and finds a significant elasticity of substitution between “high-speed” (> 10 Mbps) wired broadband and “lower speed” satellite and DSL. Grzybowski et al. (2018) analyze the European broadband subscriber data in 2014 and find that consumer valuation on the improvements from 1 Mbps (DSL) to 8 Mbps (DSL), and then to 100 Mbps (FttH, Fiber to the Home) are small, but are increasing over time. The paper also indicates that the switching cost on adapting FttH, such as installation cost and uncertainty, significantly slow the transition process even when the fiber technology is deployed by the Internet service provider. In our paper, we assess whether household willingness-to-pay (WTP) for bandwidth is substantial up to 1 Gbps, importantly covering speeds surrounding 25 Mbps—the FCC's current definition for “broadband.”

A major challenge of the aforementioned studies using market data is their limited ability to pin down how much households value speed increments, let alone increments in separate components of speed. This limitation in empirical measurements is largely due to limitations in market data; it has been difficult or even impossible to get data on broadband sales and prices at varying speeds that has sufficient and exogenous variation (Rosston et al., 2010). The limitations of market data are especially apparent when measuring the WTP for very-high bandwidth, as the variation in availability creates a selection problem. Given these challenges with market data, we chose to build on the stream of literature utilizing surveys and experiments.

Prior literature using surveys and experiments to measure the value of Internet speed generally focused on measuring bandwidth. In a late 90′s experiment, Varian (2002) allowed the research team to vary the metered price exogenously for bandwidths ranging from 8–128 Kbps (6 levels). The experiment revealed that participants have a low WTP for improved speeds within this range. Varian concluded that “the problem with broadband is not access but applications.” Rappoport et al. (2002) found a contradictory result using survey data in 2000: Broadband users no longer perceived dial-up as a substitute, as evidenced by the nearly zero cross-price elasticity. In a later study, Dutz et al. (2009) further found that from 2005–2008, the substitution pattern between dial-up and broadband becomes weaker, and the own-price elasticity of broadband becomes less elastic, indicating overall increases in the residential value of broadband. Additional studies assessed broader Internet performance factors (e.g. connection reliability, always-on feature) and the preferences of heterogeneous users (e.g. rural/urban, high/low technical ability, experienced/inexperienced Internet users) (Savage and Waldman, 2005, Savage and Waldman, 2009, Rosston et al., 2010).

A notable limitation of these studies stems from ambiguity regarding the meaning of Internet speed. Most of them regard bandwidth as equivalent to speed, but this assumption is not entirely adequate. To users, Internet speed means how quickly they can complete a given task. From this perspective, bandwidth is just one factor that determines Internet speed, as tasks involve more than just downloading and uploading files. Higher bandwidth may not necessarily lead to perceivable improvements in efficiency in many cases. In the most recent survey study of which we are aware, Rosston et al. (2010) describes the top speed level as “Blazing fast downloads and uploads. It is really great for gaming, watching high-definition movies, and instantly transferring large files.” For highly dynamic activities, such as gaming or real-time communication, it is latency, not bandwidth that determines “speed” for a typical American household (Grigorik, 2013). The WTP being captured using this description is thus a mixture of the value of both bandwidth and latency. Latency is also primarily responsible for web page loading time and a primary bottleneck of speed for several other activities, due to the significant improvements in bandwidth in the recent decade. To the best of our knowledge, our paper is the first to economically evaluate the importance of latency.

Section snippets

Internet speed and policy

Internet speed has become an integral part of setting policy. The FCC has established definitions, in Mbps, of the minimum bandwidth to be considered broadband, and another definition used to be eligible for the $4.5 billion annually from the universal service program's Connect America Fund. Yet, as discussed above and detailed below, speed is a function of both bandwidth and latency. The Connect America Fund Phase II Auction recognizes the dual nature of speed by taking into account proposed

Survey design

To estimate the WTP of the Internet speed components, we collect and analyze data from a nationwide online survey that employs repeated discrete choice experiments (DCEs). Prior work has shown that DCEs mitigate the reporting inaccuracy of stated-preference data (Carare et al., 2015). Even if hypothetical bias may potentially overestimate demand, the WTP estimation for changes in feature levels is statistically unbiased (Ding et al., 2005, Miller et al., 2011).10

Data

Our data come from ResearchNow's (RN) standing Internet panel. As we aim for 1,050 and 450 completed surveys for the latency and no-latency surveys respectively, RN makes sure that the target sample sizes are satisfied and the sample composition of responders is similar to the US census, particularly on geographical region, sex, ethnicity, and age. A qualified record requires the household respondent to be at least 18 years old, have a home Internet plan, be the primary decision-maker for the

Econometric methods

We use the conditional logistic regression model (McFadden, 1974, Greene, 2012) to estimate utility parameters and ultimately calculate the WTP. Let xijk be a vector of attributes for alternative j in choice question k that individual i faces. A linear random utility model can be written as:uijk=xijkβ+ɛijk

We interpret the errors (ɛijk) as individual idiosyncratic preference and assume that it is independently and identically distributed with type I extreme value distributions. With this

Results

Table 5 contains our parameter estimates for both the latency and non-latency surveys. In columns (1) and (4), we use the responses of the hypothetical choice experiment only. As each respondent makes eight choices, the effective sample size is eight times of the number of respondents. Columns (2), (3), (5), and (6) include the follow-up status quo questions.20

A comparison of WTP estimates to FCC point tradeoffs

With these figures in hand, we can compare our relative WTP estimates to the FCC's point tradeoffs for its CAF Phase II auction (see Tables 1 and 2). Tables 8 and 9 make these comparisons.23

Conclusions

In this paper, we measured households’ valuations of Internet connection characteristics, focusing on speed and data caps as is done in the CAF Phase II Auction scoring. We find that consumers highly value bandwidth enhancements at lower speeds, but the incremental value of bandwidth decreases rapidly. Further, household valuation of allowed data transmission (data caps) is similarly concave, but consumers appear to place a premium on unlimited data transfer. Lastly, households indicate that

References (23)

  • Broadband speed guide,

    (2016)
  • Cited by (32)

    View all citing articles on Scopus

    We thank several focus groups for their time and insight and ResearchNow for assistance in administering the surveys. We are responsible for all errors.

    View full text