Data Sources
Source of truth: the "Sources and download dates" section of
calculations/outputs/u_probabilities.md, and calculations/config.py, which records how each URL was resolved.
The calculations project uses only public aggregate data. It uses no patient data. As recorded in the generated output, all data was downloaded or re-verified on 2026-10-06.
Sources used for the computed values
| Used for | Source | Publisher |
|---|---|---|
| First names (2020), and the proxy for middle names | Names2020_FirstNames_Sex.xlsx | U.S. Census Bureau |
| Last names (2020) | Names2020_LastNames_RaceHispanic.xlsx. This file publishes race and Hispanic-origin cross-tabs as well as a total frequency column. The tool uses the total column, which the project's configuration notes was verified to equal the sum of the per-category columns. | U.S. Census Bureau |
| Last names (2010), comparison only | names_2010census.zip | U.S. Census Bureau |
| Year of birth and full date of birth | nc-est2025-agesex-res.csv (national single-year-of-age estimates) | U.S. Census Bureau |
| State | NST-EST2025-POP.xlsx (July 1, 2025 estimate) | U.S. Census Bureau |
| City | sub-est2025.csv (incorporated places and Census Designated Places, July 1, 2025 estimate) | U.S. Census Bureau |
| ZIP (via ZCTA), street line | American Community Survey 2024 5-year estimates at ZCTA level: population (B01003), household size (B11016, B25010), housing units (B25001) | U.S. Census Bureau |
| MBI | "Medicare Total Enrollment" (2024), Program Statistics | Centers for Medicare & Medicaid Services |
| Phone (share of adults in a household with a landline, 20.7%: 19.8% dual-user plus 0.9% landline-only) | NCHS Wireless Substitution survey, July to December 2024, Table 1: doi.org/10.15620/cdc/174608 | CDC National Center for Health Statistics |
| SSN / ITIN last 4 | SSA SSN Randomization policy (closed form, no data file): ssa.gov/employer/randomization.html | Social Security Administration |
Cited but not used in a computed value
- FCC 18-31, CG Docket No. 17-59, para. 3 (2018-03-22): cited in the phone notes as a named, deliberately unquantified gap (mobile number reassignment, about 35 million U.S. numbers per year). No data file is used.
Not used
- HUD USPS ZIP crosswalk: skipped because no
HUD_TOKENwas configured, so the alternate USPS-city-via-ZIP calculation for city is not part of the published numbers.
Reproducibility of the data
- The raw files are committed in the repository under
calculations/data/raw/(about 41 MB), with SHA-256 checksums inSHA256SUMS.make calculatetherefore runs offline, and an upstream change to a source file shows up as a diff. - ACS 2024 5-year estimates are used because the 2025 5-year estimates were not yet published as of the project's last documented run.
- Census name files list only names occurring 100 or more times. See Known Limitations.
Test data is separate
The labeled test cases come from a different public dataset, the ONC 2017 Patient Matching Algorithm Challenge dataset. See Test Set Methodology and Test Set Overview for its provenance and attribution.
Was this page helpful?