Definitions, sources, and methodology — firearm regulation map

What this site shows

Each layer on the map displays a single variable for a single year, across all 50 U.S. states. The full underlying dataset is a balanced state-year panel covering 1979–2024 for the law and crime variables, with shorter coverage windows for variables whose source series start later (NICS background checks from 1999; ACS demographics from 2008; RAND household ownership 1980–2016; firearm suicide / homicide / FS ownership proxy 1979–2023). Year sliders gracefully gray out states without data in that year.

The site is generated from CSV/JSON files in the same git repository. The data pipeline is in scripts/build_website_data.py, which reads the project’s balanced panels and emits docs/data/panel.json and docs/data/metadata.json. To rebuild after a data update: python scripts/build_website_data.py.

Important caveats

Background checks are not gun sales. NICS counts include some states’ permit re-checks, and a single check can cover multiple firearms. Treat NICS as a market-activity flow proxy, not a stock measure of gun owners.
The FS/S ownership proxy is mechanically correlated with suicide outcomes. It is the share of all suicides committed with a firearm, so it should never be used as an explanatory variable when the outcome is suicide. The RAND HFR estimate (1980–2016) is the more defensible ownership measure for cross-state work.
FBI rape definition changed in 2013. Pre- and post-2013 rape rates are not directly comparable.
Tufts law coding stops at 2024. Bruen-era injunctions are not generally re-coded; check the Tufts narrative report for state-specific exceptions.
2026 data are incomplete. The map currently goes through 2024 because the latest full-year FBI crime release covers 2024.

Manipulations applied to the raw data

Variable family	What was done	Why
Granular crime (homicide, robbery, ...)	The OpenCrime feed double-listed North Carolina 2022 and omitted North Dakota 2022. The second NC 2022 row (population ≈ 779k) was reassigned to ND.	Logged in `data/processed/crime_repairs_log.csv`. Applied identically in the website build to keep all granular crime variables balanced.
Pre-ACS demographic shares (1990–2007)	Reconstructed annual non-Hispanic race shares, male share, and age-group shares from Census ASRH (1990–1999), the 2000s intercensal all-data file, and PEP (2010–2024).	ACS 1-year tables are only available from 2008. The reconstruction matches Census published denominators within rounding.
Bachelor’s+ share before 2008	Linearly interpolated within state from Census Statistical Abstract Table 229 anchors (1990, 2000, 2008) and ACS 2008–2024 actuals.	Single-year educational attainment estimates are not consistently available before ACS.
SAIPE income / poverty	Missing 1990–1992 and 1994 values are linearly interpolated within state.	SAIPE skipped those years; interpolation preserves a balanced panel.
Per-capita income, household income	BEA per-capita personal income and SAIPE median household income are deflated to 2024 dollars using CPI-U.	Real-terms comparability across the long panel.
Firearm suicide rates / total suicide rates	Computed on the fly as `firearm_suicides / total_population × 100000` (and similarly for total suicides) from the firearm-suicide-share dataset.	Source file ships counts and the FS/S ratio but not population-normalized rates.
Law-category indices	Derived as simple sums of the relevant Tufts binary indicators (background checks, carry restrictions, banned weapons, buyer restrictions, DV prohibitors, storage / GVRO).	Group-level totals are easier to read on a map than 72 individual indicators. The category groupings are listed in `scripts/build_website_data.py`.

All variables — State view (1979–2024)

Variables are grouped by category. Click a source link to open the original data provider in a new tab.

Loading state variables...

All variables — County view (2009–2024)

The county view exposes a different (smaller) set of variables: county crime from Jacob Kaplan's UCR aggregations, county demographics from ACS 5-year, county economy (SAIPE income/poverty + ERS unemployment + BEA per-capita income), and state firearm laws plus state firearm mortality joined down to every county in a state-year.

Loading county variables...

Data sources cited on this site

Tufts CTSI — State Firearm Laws Database (1976–2024) · coded by Michael Siegel et al., 72 binary indicators across 11 categories.
FBI Uniform Crime Reports / Crime Data Explorer · via OpenCrime extraction (state annual counts and rates).
FBI NICS background checks · via Data Liberation Project.
RAND TL-354 — State-Level Estimates of Household Firearm Ownership (1980–2016) · Schell, Peterson, Vegetabile, Scherling, Smart, Morral.
Firearm suicide / homicide v2 dataset · long-run state firearm suicides, total suicides, firearm homicides, and the FS/S ownership proxy (1949–2023).
U.S. Census Bureau Population Estimates Program · ASRH, intercensal all-data, PEP files.
Census SAIPE · small-area income and poverty estimates.
U.S. Census Bureau ACS 1-year detailed tables (2008–2024).
FRED (Federal Reserve Bank of St. Louis) · BLS LAUS unemployment, BEA per-capita personal income (state).
Census Statistical Abstract Table 229 · pre-ACS educational attainment anchors.
Jacob Kaplan's Concatenated Files: Offenses Known and Clearances by Arrest, 1960–2024 · openICPSR project 100707, V22. Source for county-level UCR crime counts.
USDA Economic Research Service county-level data sets · mirrors BLS LAUS county unemployment.
BEA CAINC1 county personal income tables · per-capita personal income at the county grain.

How to reproduce

Clone this repository.
Install Python with pandas, numpy, and openpyxl.
State panels: python scripts/build_firearms_panel.py (rebuilds the original data/processed/*.csv from raw inputs), then python scripts/audit_panels.py and python scripts/augment_panels.py.
County panel (Phase 1+2a): python scripts/build_county_panel.py. Auto-fetches ACS 5-year via the Census API (cached) and reads PEP/SAIPE/ERS/BEA flat files from data/county/.
County crime (Phase 3): manually download Jacob Kaplan's openICPSR project 100707 V22 and unzip into data/county/kaplan_offenses/, then run python scripts/build_county_crime.py. Re-run build_county_panel.py to merge the crime layer in.
Website data files: python scripts/build_website_data.py (state) and python scripts/build_website_county_data.py (county per-year files).
Preview: python -m http.server 8000 -d docs then open http://localhost:8000.

The full prose log of every data source and manipulation lives in data_appendix.md at the repository root.

Generated at —