Dataset #1: United States Broadband Usage Percentages
Dataset Overview and Geographic Insights into the Digital Divide
Our primary dataset contains broadband availability and usage data for counties across the United States during November 2019 and October 2020.
The specific data fields are:
- ST: The 2 letter abbreviation of the state.
- COUNTY ID: A 4 to 5 digit code used to represent the county (last 3 digits) and the state (first digit or first 2 digits).
- COUNTY NAME: The name of the county.
- BROADBAND AVAILABILITY PER FCC: Percentage of the population in each county with access to fixed terrestrial broadband at speeds of 25 Mbps/3 Mbps (as of the end of 2019).
- BROADBAND USAGE: Percentage of the population in each county using broadband internet at the specified speeds (October 2020).
The dataset illuminates information on infrastructure and behavior in internet access, and has the potential to reveal narratives about the digital divide—the gap created by disparities in access to the internet and digital devices. More specifically, the dataset provides insights into geographic disparities in broadband availability and usage, which can inform policies aimed at expanding coverage and evaluating the effectiveness of existing programs. By comparing reported broadband availability with actual usage, the dataset also raises questions about the accuracy of broadband data, potentially highlighting instances where coverage may be overstated.
Beyond technical details, the dataset offers a lens into broader social and economic issues: areas with low broadband adoption often face barriers in education, healthcare, employment, and civic engagement. Thus, this dataset not only provides technical insights about internet access but also unveils larger patterns of inequality, opportunity, and social inclusion across the United States.
Data Generation, Sources, and Funding Background
The dataset was created and funded by Microsoft as part of its Airband Initiative, a corporate program aimed at closing the rural broadband gap in the United States. As a tech giant, Microsoft’s interest in broadband infrastructure and digital access means that the framing of the dataset may align with corporate priorities, potentially introducing bias into the focus areas and interpretation of the data.
The dataset was created by integrating data from the U.S. Census, the Federal Communications Commission (FCC), and Microsoft’s own service data. Estimates of fixed broadband access primarily rely on the FCC’s annual report. The FCC, a U.S. government agency that regulates communications, collects broadband availability data via Form 447, a mandatory filing that broadband providers must submit to indicate whether service is available at their corresponding census blocks. The FCC uses this form to track the deployment and availability of broadband nationwide. A census block is considered “served” if the form indicates that any single location within a census block is served by a broadband provider, regardless of whether all areas are covered. This methodology may overestimate broadband coverage, as not every household or individual in that block is guaranteed access. Nevertheless, these data remain valuable for its fine granularity and consistency in the unit of measurement with other datasets we will use.
To estimate actual broadband usage, Microsoft also uses data from multiple Microsoft services. When a device connects to a Microsoft service, such as by downloading a Windows update or accessing Microsoft software, Microsoft measures the size of the data sent and the time it takes to estimate the throughput speed of a device. Devices that achieve download speeds of at least 25 Mbps—the FCC’s minimum threshold for broadband—are counted, and the number of such devices is compared with census data on households to estimate actual broadband adoption. This method effectively captures the “usage gap,” where infrastructure exists but broadband is underutilized due to barriers like cost or digital literacy. However, it may underestimate usage in households that rely on non-Microsoft devices and services.
Limitations and Ontological Framing of the Dataset
Several important factors are omitted from the dataset, which can limit its ability to fully represent broadband access. For example, tribal communities that are not federally recognized are excluded from the FCC’s reports, potentially leading to an underrepresentation of these populations. The dataset also lacks income data about each county, making it impossible to draw direct correlations between broadband access and wealth inequality. Without demographic data such as age, education level, and employment status, the dataset cannot account for these factors that likely influence internet access and usage.
The ontology in our dataset significantly influences the way we interpret broadband access in the U.S. Specifically, the dataset categorizes broadband access by state and county:
- Segmenting by state might highlight differences based on political leadership and state policies, potentially reinforcing the ideological view that broadband access is largely shaped by government intervention. For instance, states with progressive policies may appear to have better access, while conservative states could seem to lag behind.
- County-level breakdowns may reveal disparities between urban and rural areas, likely highlighting rural counties as needing more investment, which could reinforce the idea that rural areas are the primary focus of broadband policy. However, this may overlook other important factors such as income inequality or local government efforts to improve access in urban areas.
The decision to structure the dataset by state and county can shape narratives around political, geographic, and socio-economic factors–potentially reinforcing or overlooking various ideological assumptions about broadband access.
Dataset #2: Disparities in Internet Deals
Dataset Overview and Demographic Insights into the Digital Divide
Our dataset is derived from The Markup’s investigation, “How We Uncovered Disparities in Internet Deals,” which examines broadband pricing and availability across 38 U.S. cities. It includes detailed data on over 800,000 residential addresses, focusing on five major Internet Service Providers (ISPs): AT&T, Verizon, EarthLink, CenturyLink, and Frontier. The data allows for a detailed analysis of how broadband service offerings (price, speed, data caps, and promotional pricing) vary by geographic location, particularly influenced by socioeconomic and racial demographics.
Key pieces of information include:
- Available Plans: The internet plans offered at each address, including pricing, download/upload speeds, data caps, and any promotional pricing.
- Address-Level Granularity: Each entry is tied to a specific residential address, facilitating precise geospatial analysis.
- Plan Metadata: Details such as plan type (fiber or DSL), new customer offers, and price variations over time (e.g., introductory vs. long-term pricing).
- Neighborhood Demographics: Enriched with U.S. Census data, this includes median income, racial composition, and the percentage of households with broadband subscriptions in the area.
The dataset sheds light on disparities in broadband access, particularly the digital divide, where some neighborhoods—especially low-income and communities of color—are offered slower internet speeds for the same or higher prices compared to wealthier, predominantly white areas. It highlights discriminatory pricing patterns, where underserved areas are given worse deals in terms of service quality and price. This data also reveals how historical redlining continues to impact broadband access, with historically marginalized communities disproportionately receiving lower quality broadband for the same cost.
Data Generation, Sources, and Funding Background
The dataset was created through The Markup’s collaboration with the Associated Press, using automated tools to collect broadband service offers from the websites of the ISPs. This was done by simulating the process of shopping for broadband at over 800,000 residential addresses. The data collection methodology draws from multiple sources:
- ISP Websites: Information on broadband pricing, speeds, and availability was gathered from the lookup tools provided by AT&T, Verizon, EarthLink, CenturyLink, and Frontier. These websites allowed the research team to access broadband offers at a large scale.
- Census Data: The dataset was enriched with demographic data from the U.S. Census, including information on neighborhood income levels, racial composition, and broadband adoption.
- Historical Redlining Data: The research incorporated data from the University of Richmond’s Mapping Inequality project, which provided historical redlining maps that classified neighborhoods based on their perceived investment risk.
The primary funding and organizational support for this project come from The Markup, a nonprofit journalism organization that focuses on the intersection of technology and society. They were responsible for compiling, analyzing, and interpreting the data.
Limitations and Ontological Framing of the Dataset
While the dataset provides rich insights into broadband pricing and availability, there are several important limitations:
- ISP Intent or Internal Policy Data: The dataset reveals what ISPs offer but does not provide insight into why these offerings exist. The intent behind ISP pricing decisions—such as whether certain areas are intentionally underserved—is not captured. Thus, the dataset cannot prove discriminatory practices or explain the reasoning behind disparities in broadband service offerings.
- Customer Experience Data: The dataset focuses on the plans offered by ISPs but lacks data on customer choices, actual service quality, or customer satisfaction. For instance, a plan may be offered but still be financially inaccessible due to cost, digital literacy, or hidden fees. This missing information limits our understanding of the real-world usability of these broadband options.
- Geographic Coverage: The dataset is limited to five ISPs (AT&T, Verizon, EarthLink, CenturyLink, and Frontier) and only covers 38 cities, with no rural coverage or data on other smaller or regional providers. As a result, the findings are not fully generalizable to the entire U.S. population, especially in rural areas, where different providers may operate, and broadband access may differ significantly.
- Dynamic Nature of Broadband Offers: The data represents a snapshot in time (collected between April and October 2022), and broadband pricing and service availability can change frequently. Since ISPs regularly update their pricing and promotional offers, the dataset might not capture long-term trends or seasonal fluctuations in broadband plans.
The ontology of the dataset is structured by address-level granularity, which provides a detailed view of service offers in specific geographic locations. Each address is linked to neighborhood-level demographics and enriched with data about broadband infrastructure, plan pricing, and historical factors like redlining. While this granularity allows for precise analysis of pricing disparities, it may not fully account for broader systemic factors that affect broadband access, such as unadvertised fees, network congestion, or community efforts to improve infrastructure.
The dataset also focuses on broadband offers and prices rather than actual broadband usage or adoption behavior, leaving out key insights into how these factors play out in practice across different socioeconomic and racial groups.
Dataset #3: FCC State Broadband Funding
Dataset Overview and Policy-Driven Insights into the Digital Divide
The FCC’s RDOF dataset includes information on federal broadband infrastructure investments awarded through Auction 904. It contains fields such as total assigned support, number of service locations, and winning bidders for each U.S. state and census block. This dataset can illuminate where and how the federal government is investing in broadband infrastructure, especially in rural and underserved areas. It helps identify geographic patterns in funding, which states or regions were prioritized, and the scale of support allocated to specific providers.
Key pieces of information include:
- State: Indicates which U.S. state received broadband funding through the RDOF auction.
- Total Assigned Support over 10 Years: The total amount of federal funding awarded to support broadband buildout over a ten-year period.
- Number of Locations Assigned: The number of locations (homes or businesses) that providers committed to serve with broadband in each state.
- Winning Bidders: The internet service providers selected to carry out the broadband expansion in the funded areas.
Data Generation, Sources, and Funding Background
The data was generated by the Federal Communications Commission (FCC) based on the outcomes of its 2020 Rural Digital Opportunity Fund (RDOF) Auction 904, a reverse auction funded through the Universal Service Fund (USF). In this auction, internet service providers competed to receive government subsidies by bidding to serve eligible areas lacking broadband. The FCC determined eligibility using internal broadband coverage maps and census block data, while the auction results were recorded based on winning bids and proposed service areas.
Limitations and Ontological Framing of the Dataset
The dataset cannot reveal whether broadband was actually deployed as promised or the quality, speed, and affordability of the service that was eventually delivered. It also does not include demographic data such as income, race, education level, or digital literacy, which are crucial for understanding broadband adoption and equity. Its ontology is infrastructural and economic – it frames broadband access in terms of financial commitment and geographic coverage, not in terms of user experience or social impact. If this dataset were the only source, it would leave out critical insights into who is benefiting from broadband expansion, how it’s being used, and why certain communities may still lack meaningful access despite infrastructure investments.