How this site is built

Sources

ARCOS county shipments

County-year opioid shipment aggregates 2006–2014 from the Griffith et al. Data in Brief ARCOS dataset. View at Mendeley Data. Released under CC BY 4.0.

DEA Diversion Control

Annual counts of DEA enforcement actions are pulled from the Federal Register API — specifically, NOTICE documents published by the DEA whose titles match a registrant-action disposition: Final Orders, Revocations of Registration, Immediate Suspension Orders, Orders to Show Cause, Settlements, and Admonitions. Classification is regex-based on notice titles; post-2011 the Federal Register switched to an umbrella "Decision and Order" title convention, which means later-year type breakdowns are coarser than earlier years. Counts reflect the year of Federal Register publication, not the year of the underlying conduct. Criminal prosecutions (which proceed through DOJ, not DEA's administrative track) are not included. Source: federalregister.gov/api. Public domain (17 USC §105).

CDC WONDER

Overdose deaths by county-year from a CDC WONDER Underlying Cause of Death 1999-2020 interactive UI scrape. View at wonder.cdc.gov. The scrape runs one state/DC query at a time for 2006-2014, using the WONDER Drug/Alcohol Induced Causes D1-D4 macro and ICD-10 codes X40-X44, X60-X64, X85, Y10-Y14. Counts of 9 or fewer are suppressed under 42 USC 242m(d) and rendered <10, never zero. Counts of 10-20 are publishable as raw deaths, but CDC flags their rates as statistically unreliable.

Joins

The pipeline builds a canonical FIPS × year grid from Census PEP population estimates, then LEFT JOINs each cleaned source. The result lives at data/joined/master.parquet and is the input to every emitted artifact.

Caveats

• ARCOS covers 2006–2014 only. Later years are not in this dataset.
• CDC suppression hides counts of 9 or fewer deaths in a county-year — the map renders these as <10, never zero. Rates based on 10-20 deaths are flagged by CDC as statistically unreliable even though the raw death counts are publishable.
• Pill counts are in DEA "dosage units," not individual pills; a 100mg tablet counts as one unit regardless of strength.
• DEA enforcement totals reflect Federal Register publication dates, not the dates of the underlying conduct. Final orders routinely lag the initiating Immediate Suspension Orders by 6–24 months.

Licenses

• Code: Apache 2.0 — View license
• Fonts: SIL OFL 1.1 (Space Grotesk, Inter) — View license
• Data: Each source's upstream license applies; see above.

Access dates

The data bundle is refreshed weekly by .github/workflows/build-data.yml. The build date is stamped in the site footer.