Broadband Coverage in the US: When Sources Disagree

Categories: Big Data, Quantitative

Tags: , , , ,

Every summer, the Biocomplexity Institute’s Social and Decision Analytics Division’s Data Science for the Public Good (DSPG) Young Scholars program draws university students from around the country to work together on projects that use computational expertise to address critical social issues faced by local, regional, state or federal governments. The students conduct research at the intersection of statistics, computation, and the social sciences to determine how information generated within every community can be leveraged to improve quality of life and inform public policy. The program, held at the University of Virginia’s Arlington offices, runs for 10 weeks for undergraduate interns and 11 weeks for graduate fellows who work in teams collaborating with postdoctoral associates and research faculty from the division, and project stakeholders.

The 2019 cohort conducted nine research projects, and their methodologies and discoveries will be presented at MethodSpace over the next three weeks as part of our examinations of Methods In Action. The descriptions of the projects were penned by the students themselves, and their names, mentors and sponsors appear under the DSPG logo in the text.

ABSTRACT - Broadband Data Validation: Comparing U.S. Broadband Coverage
USDA would like to know whether broadband development has an effect on rural prosperity, as measured by changes in property values and quality of life indicators in rural communities. This DSPG project has two parts that will inform the overall project in the Social and Decision Analytics Division. The first is evaluating and benchmarking the quality of the Federal Communication Commission Form 477 data on broadband availability and subscription rates. The second is identifying what factors other than broadband predict property values.

We Don’t Know How Many Americans are Without Broadband

The Federal Communications Commission (FCC) provided $1.49 billion in 2018 to fund broadband development in rural areas. The same year, the United States Department of Agriculture (USDA) created the ReConnect Program, providing $600 million in funding for broadband improvement in rural areas.

Despite these programs and the agencies’ efforts to expand broadband coverage, internet at broadband speeds—defined as 25 megabytes per second download and 3mbps upload—is still not available to many Americans. However, estimates differ widely on how many individuals are without access to broadband and thus limited in their ability to participate in today’s increasingly online world.

To investigate where US broadband coverage may not be good, our Data Science for Public Good team set out to investigate three broadband data sources: the FCC, the American Community Survey (ACS), and Microsoft Airband data. Our aims were to understand the extent of coverage according to each dataset, to examine discrepancies between the FCC data and its alternatives, and to address both aims with particular attention to rural areas.

Typically, broadband availability estimates come from the FCC. The FCC is a government agency data source on broadband coverage and subscriptions that informs decision-making and funding policies. However, the dataset has known limitations. The FCC reports a census block group  (a geographical cluster containing between 600 and 3,000 people) as having broadband available even if only one subscriber has it or if the provider is not currently providing coverage but could feasibly start doing so in an area within a standard service interval. It also collects advertised rather than actual internet speeds. These factors contribute to potential coverage overestimation.


Students: Sarah McDonald, Raghav Sawhney, Kateryna Savchyn,
Mentors: Teja Pristavec, Josh Goldstein, Stephanie Shipp
Sponsor: United States Department of Agriculture Economic Research Service

Broadband access is associated with beneficial outcomes like improved firm productivity, increased employment opportunities, higher political participation, and better education outcomes. Given that broadband facilitates participation in an increasingly online economy, a better understanding of the state of US broadband coverage is needed. Accurately identifying areas with low broadband access has implications for broadband policies and funding and is particularly important for rural communities struggling to be competitive.

Three Datasets Measuring Broadband Coverage

We used three public data sources in comparing US broadband coverage, each available at different geographic levels, reported by a different party, and with different limitations.

The FCC collects broadband availability and subscription data from providers and makes data available across different years at census block group, census tract, and county levels. For better comparisons with other sources, we used 2015 data. As noted, providers can report broadband as available at the census block level to FCC even if they are not yet providing access but could, or if only one household has access.

Our second source, the US Census Bureau’s ACS, is an annual, statistically designed, and nationally representative US household survey. It provides estimates and margins of error on population sociodemographic characteristics and select topics, including internet access. Contrary to the FCC data, the ACS relies on self-reports. Broadband estimates became available for the first time at the census block level in its 2013-2017 (5-year) data, which we used in our analyses. ACS’ limitation is that its survey question on internet access does not define broadband, so reporting whether or not they have an internet subscription that provides broadband speeds is left to the survey responder.

Finally, the Microsoft Airband Initiative data is neither provider- nor consumer-reported; instead, it comes from Microsoft’s one-time initiative to collect data on broadband coverage using customer access. Microsoft analyzed its server logs when electronic devices downloaded Microsoft Windows and Office updates, accessed Microsoft’s Bing search, or used Xbox gaming consoles; we cannot be certain that this is a representative sample of services that fully captures broadband usage. The company aggregated their 2018 data at the county level.

Capturing Coverage and Discrepancies

Our challenge in comparing broadband coverage across these three data sources—each using different geographic levels and methods of obtaining information, was constructing indicators that capture coverage in similar ways.

We calculated FCC broadband coverage as the proportion of the population with access to at least one provider offering at least 25mbps download speed. We used FCC provider-reported maximum advertised downstream speeds coupled with ACS population estimates. Similarly, we calculated ACS broadband coverage as the proportion of the population self-reporting access to a broadband connection (excluding satellite and cellular). Our Microsoft broadband coverageindicator captured the proportion of population that used Microsoft services at 25mbps broadband download speed, according to Microsoft server logs. We calculated each metric for every available geographic level.

We constructed reported coverage discrepancy measures to examine whether and how broadband access information differs across data sources. Our reference point was FCC coverage as it is the source used in official statistics and policymaking. We constructed FCC-ACS and FCC-Microsoft broadband coverage discrepancy measures indicating the percent discrepancy in reported coverage between the dataset pairs (i.e., percent FCC coverage minus percent ACS coverage, and percent FCC coverage minus percent Microsoft coverage). We calculated each metric at each geographic level available.

To better understand disadvantage in broadband access for rural versus urban communities, we also classified geographies according to USDA Rural-Urban Continuum Codes (RUCC). RUCC codes classify metropolitan counties by the population size of their metro area, and nonmetropolitan counties by degree of urbanization and adjacency to a metro area. We grouped rural and urban codes to construct an indicator of whether a geography was a metro or nonmetro area.

Millions Without Broadband, But How Many and Where?

Figure 1 shows the overall county distribution of broadband coverage according to FCC provider reports, ACS consumer reports, and Microsoft server logs.

Figure 1: County distribution of broadband access by data source.

As Figure 1 shows, the FCC suggests an overall higher broadband coverage within counties compared to ACS and Microsoft data. The FCC receives provider-reported data on advertised coverage from providers, which may introduce a positive coverage bias if advertised estimates do not match actual estimates, and it has other known data quality issues potentially leading to overestimates. Accordingly, the FCC claims that broadband is not available to 25 million Americans, whereas the Microsoft Airband Initiative—the data source which suggests the lowest broadband coverage—finds that around 163 million Americans do not use internet at 25 mbps. Given that Microsoft’s data is based on server logs of Bing, Microsoft Update, and XBox use, this sample likely excludes non-Windows users, and may underreport broadband coverage. ACS shows coverage between the FCC and Microsoft extremes, but similarly to Microsoft, it suggests that US broadband coverage is lower than FCC reports. 

Numbers Aside, Urban Areas Doing Better than Rural Ones

Even though the datasets differ in their estimates of US broadband coverage, they all show an urban-rural disparity. Figure 2 visualizes broadband coverage reports by urbanicity and shows that on average and across all three datasets that metro areas have higher broadband coverage compared to non-metro regions.

Figure 2: County distribution of broadband access by data source and urban status.

Exploring Dataset Discrepancies Through an Interactive Dashboard

We developed a dashboard to illustrate our discrepancy findings. Users can select a state of interest, choose a geography, and subset visualizations into urban and rural areas. Each visualization overlays major cities in each state. At each geographic level, users can view all available data for the selected area by hovering over a county, tract, or block group. Figure 3 is an example from the dashboard, showing the FCC-ACS broadband coverage discrepancies for Alabama. Our maps show that the range of discrepancies tends to be larger in rural versus urban areas.

Figure 3: Alabama FCC-Microsoft broadband coverage discrepancy map at county level.

Discussion and Next Steps

Discrepancies in reported broadband coverage exist but regardless of data source, urban areas appear to have higher coverage than rural ones. These report discrepancies are larger  in rural than urban areas. Our next step is to develop heuristics for FCC data reliability and provide tools for identifying where FCC data should be supplemented or improved using the three data sources to obtain model-based broadband access predictions based on urbanicity and sociodemographic information.

Another line of inquiry could investigate cellular coverage. Although cell phone internet access cannot replace a fixed broadband connection—cellular networks are slower, have higher latency or delay before data transfer, and are optimized for flexibility rather than continuous coverage—cellular access could supplement or partially replace a fixed connection in some use cases. Having coverage data, broadband or cellular, that is accurate and high quality is needed to ensure that decision-makers direct broadband development resources to areas that need it most.

Leave a Reply