CRSS2024CSV.zip

File Identity

CRSS2024CSV.zip is the 2024 CSV package for NHTSA's Crash Report Sampling System. CRSS is a redesigned successor to the retired National Automotive Sampling System General Estimates System (NASS GES), first published for 2016 data. Unlike FARS — which is a census of every U.S. traffic fatality — CRSS is a probability sample of roughly 50,000 police crash reports drawn from the more than 6 million police-reported crashes that occur on U.S. trafficways each year.^1,5

Because CRSS uses a stratified, three-stage probability design, each record carries sampling weights. Multiplying records by those weights projects the sample back to a national estimate. Unweighted record counts do not represent actual U.S. crash totals.^4,5

CRSS vs. FARS: Key Differences

CRSS and FARS are complementary datasets maintained by the same agency, but they serve different research purposes. FARS is a census of fatal crashes; CRSS is a weighted sample of all reported crash severities.^1,5

What the Package Contains

The ZIP contains separate CSV tables for crash-level, vehicle-level, person-level, and factor-level records. The three primary tables are accident.csv, vehicle.csv, and person.csv. The sampling weight fields in accident.csv (PSU, PSUSTRAT, WEIGHT) must be carried through any analysis that joins to vehicle or person records.^3,4

Sampling Weight Fields

CRSS is a probability sample. Every analysis that uses CRSS to make statements about national crash totals must apply the sampling weights. The three fields below appear in accident.csv and must be used with survey-analysis procedures — not standard regression or simple counts.^4,5

How Daily Driver Advocate Uses It

Daily Driver Advocate primarily uses FARS for fatal crash and fatality statistics, where an exact census count is required. CRSS is the primary source when a report covers non-fatal crash patterns — such as injury crash rates, distracted driving estimates, or crash types involving pedestrians and cyclists.^1,3

Common Research Uses

CRSS is the appropriate starting point when a research question requires national crash estimates that include non-fatal incidents — the majority of police-reported crashes that FARS does not capture.^1,3,5

Important Limitations

• CRSS is a probability sample, not a census. Records must be multiplied by sampling weights to produce valid national estimates. Unweighted record counts do not represent actual U.S. crash totals.
• CRSS fatal crash estimates are lower than FARS fatal crash counts because FARS is a complete census of U.S. traffic fatalities while CRSS is a sample. For fatal-only analysis, FARS is the authoritative source.
• The 2024 file may be released as an Annual Report File (ARF) — a preliminary version subject to correction — before the final dataset is published. Check the NHTSA download directory for the current file status.
• CRSS covers crashes on U.S. public trafficways. It does not include crashes on private property or in U.S. territories.
• Approximately 27 data elements are imputed for missing values. Imputed records are flagged; researchers who need to exclude imputed data should consult the Coding and Validation Manual.
• Name, address, license number, and precise crash location are not included in the public CRSS files to protect individual privacy.

Suggested Citation

For the raw dataset, cite NHTSA as the issuing agency and link to the official file. For this explanatory card, cite Daily Driver Advocate as the independent research project that documented the source.