Important: what GUSRFI measures and does not measure
The Gera US Road Fatality Index (GUSRFI) is a statistical risk-context signal. It measures the rate of road crash fatalities relative to population and vehicle-miles traveled, smoothed over 3 years. It does NOT measure road conditions for a specific route, individual driver risk, or predict insurance premiums. High GUSRFI reflects historically higher area-level fatality rates — useful for context when assessing insurance needs or planning routes, but not a forecast of any individual outcome.
What is the GUSRFI?
The Gera US Road Fatality Index (GUSRFI) ranks all 50 US states and DC on a 0–10 scale for road fatality risk context. Higher scores indicate states where road crash fatalities per capita and per vehicle-mile traveled are historically higher, based on 3 years of NHTSA FARS data (2021–2023). The index also covers 350 high-fatality counties using a per-100k-only variant (per-county VMT is not available key-free).
- NHTSA FARS 2021-2023 — national annual crash CSVs, FATALS field, aggregated by STATE and COUNTY.
- Census PEP 2021-2023 — state and county population estimates, averaged over 3 years.
- FHWA VM-2 2023 — annual vehicle-miles traveled by state (grand total, millions VMT).
The GUSRFI formula
Step 1 — 3-year average fatalities:
avg_fatals = (fatals_2021 + fatals_2022 + fatals_2023) / 3
Step 2 — 3-year average population:
avg_pop = (pop_2021 + pop_2022 + pop_2023) / 3
Step 3 — per-100k and per-VMT rates:
per_100k = (avg_fatals / avg_pop) × 100,000
per_100m_vmt = (avg_fatals / vmt_millions_2023) × 100
Step 4 — min-max normalise each signal:
n_100k = (per_100k − 5.659) / (24.943 − 5.659)
n_vmt = (per_100m_vmt − 0.653) / (1.831 − 0.653)
Step 5 — weighted composite and scale to /10:
raw = 0.6 × n_100k + 0.4 × n_vmt
GUSRFI = raw × 10 [rounded to 1 decimal]
County GUSRFI (per_100k only, clamped):
GUSRFI = clamp(n_100k × 10, 0, 10)
Why 0.6 / 0.4 weighting? The per-population rate (per_100k) is the primary public-health metric used by NHTSA, CDC, and WHO for road safety comparisons. The per-VMT rate adjusts for driving intensity — states with more miles driven are not the same as states with higher fatality risk per mile. The 60/40 split gives majority weight to the internationally standard per-population measure while adding the VMT exposure context.
Why 3-year smoothing? Annual FARS counts can fluctuate due to weather events, data collection timing, and small denominators. A 3-year average reduces noise and gives a more stable signal of underlying risk context.
National summary (2021-2023)
| Metric | Value | Source |
|---|---|---|
| Road fatalities — 2021 | 43,230 | NHTSA FARS 2021 |
| Road fatalities — 2022 | 42,721 | NHTSA FARS 2022 |
| Road fatalities — 2023 | 41,025 | NHTSA FARS 2023 |
| Annual average (3yr) | 42,325 | Gera calculation |
| US average per 100k (3yr) | 12.695 | FARS + Census PEP |
| Highest GUSRFI state | Mississippi — 9.9/10 | Gera GUSRFI |
| Lowest GUSRFI state | Massachusetts — 0/10 | Gera GUSRFI |
Data sources and licence
1. NHTSA FARS
Contains public sector information published by NHTSA and licensed under the US Government public domain. Source: NHTSA Fatality Analysis Reporting System (FARS) — annual national CSV ZIPs 2021-2023 (FARS 2021-2023, published 2024-2025).
Fatality Analysis Reporting System — NHTSA's annual census of fatal traffic crashes on US public roads. Each row in accident.csv is one crash; FATALS is the number of people killed. STATE is a 2-digit FIPS code; COUNTY is 3-digit within state. No API key or registration required. US Government public-domain data.
2. Census Bureau Population Estimates
Contains public sector information published by US Census Bureau and licensed under the US Government public domain. Source: US Census Bureau Population Estimates (NST-EST2023, co-est2023) (Census PEP 2021-2023, published 2024).
Population Estimates Program (PEP) annual estimates for states and counties 2020-2023. NST-EST2023 for states (SUMLEV=040); co-est2023 for counties (SUMLEV=050). Both are flat CSV downloads — no API key required.
3. FHWA VM-2 Annual VMT
Contains public sector information published by FHWA and licensed under the US Government public domain. Source: FHWA Highway Statistics VM-2 — Annual VMT by State (2023) (FHWA VM-2 2023, published 2024).
Highway Statistics VM-2 table — annual vehicle-miles of travel by state, broken out by functional classification (rural/urban). Column 18 is the grand state total in millions of vehicle-miles. Available as xlsx, no key required. The 2023 file is the most recent published (as of June 2026).
Reproducing the GUSRFI
- Download FARS ZIPs for 2021, 2022, 2023 from NHTSA open data. Extract
accident.csvfrom each. - Sum
FATALSbySTATEfor each year (Python:csv.DictReaderor pandasgroupby). - Download NST-EST2023-ALLDATA.csv. Filter
SUMLEV=040. UsePOPESTIMATE2021/2022/2023. - Download vm2.xlsx. Open the "A" sheet; grand-total VMT is column 18 (0-indexed 17), state FIPS is column 19 (0-indexed 18).
- Apply the 5-step formula above. Normalisation bounds are fixed at the values computed from the 2021-2023 data (min_100k=5.659 MA, max_100k=24.943 MS, min_vmt=0.653 MA, max_vmt=1.831 SC).
- Verify Mississippi: per_100k=24.943, per_100m_vmt=1.793 → GUSRFI=9.9. Verify Massachusetts: per_100k=5.659, per_100m_vmt=0.653 → GUSRFI=0.0.
Exclusions and honesty note
- No route-level data. GUSRFI is a state and county-level area index. It does not identify specific road segments, junctions, or routes with high risk.
- No per-county VMT. FHWA's VM-2 table is state-only; county-level VMT is not published in key-free bulk form. County GUSRFI therefore uses per-100k only, clamped to [0, 10].
- FARS 2024 not yet used. The 2024 FARS national ZIP was confirmed available at the same URL pattern but the Census PEP 2024 state estimates and the FHWA VM-2 2024 were not yet published in key-free flat-file form at time of computation (June 2026). Gera will update the index when all three annual series are consistently available.
- No individual driver risk. GUSRFI is an area-level statistical signal. Individual crash risk depends on driver behaviour, vehicle type, time of day, weather, and many factors not captured in the index.