National and state averages suggest most of the country is online. County-level data tell a different story — and show that whether a county's households are actually connected tracks economic conditions far more closely than it tracks broadband infrastructure. This analysis works across 3,102 U.S. counties from a public IMLS / BroadbandNow extract to map where the gap is and what predicts it.
Takeaway: Across all 3,102 counties, poverty is the strongest correlate of home-broadband adoption (r = -0.65), more than double the correlation of formal broadband availability (r = +0.30). The digital divide in this data is mostly an affordability and adoption problem, not only a coverage problem.
- County adoption spans a 70-point range (25.7% to 95.5%; median 73.7%) that national and state averages hide.
- Economics beats infrastructure as a predictor. Poverty (r = -0.65) and SNAP receipt (r = -0.57) correlate with adoption far more strongly than formal availability (r = +0.30) or provider count (r = +0.22).
- Availability is not adoption. Of the 1,044 counties with formal availability at or above 90%, mean household adoption is still only 76.1%, and the lowest is 39%.
- The gap concentrates in the South and in high-poverty counties. Mean county adoption is 69.2% in the South vs. 79.2% in the Northeast; it falls 15 points from the lowest-poverty to the highest-poverty quartile.
- 01_Project_Summary.md — a short, plain-language overview of the problem, findings, and what they imply (about a 3-minute read).
- 02_Project_Report.md — the full write-up: data, methods, metric definitions, figure-by-figure evidence, limitations, and references.
- 03_Analysis_Notebook.ipynb — the executed analysis workflow with inline figures.
- 04_Source_Code.py — the annotated script that reproduces every statistic and figure.
An interactive Tableau version of the geographic story is available on Tableau Public: America's Broadband Problem.
County-level public extract blending the IMLS Indicators Workbook (ACS 5-year 2014–2018 estimates, with BroadbandNow and BLS inputs) and the BroadbandNow Open Data Challenge dataset. See data/README.md for provenance, variables, and cautions. Findings are descriptive patterns from observational data, not causal estimates or current-condition claims.