A Field Guide to U.S. Crash Data: FARS, CRSS, MCMIS and the Causation Studies
Each national crash number comes from a system built to answer one question. A guide to the census, the sample, the carrier files, the one-time causation studies and the compilations that repackage them.
Texas recorded 608 deaths in commercial motor vehicle crashes in 2024 by its own count and 720 deaths in large-truck crashes by the federal one162. Neither number is wrong. Each U.S. crash data system was built for one purpose: to count every death, to estimate the much larger pool of police-reported crashes, to track motor carriers, or to study why crashes happen. The numbers carry those purposes with them. This guide covers who publishes each source, what it counts and leaves out, how fast it arrives, and the trap each one sets for reporters.
The sources at a glance
| Source | Publisher | Covers | Watch out for |
|---|---|---|---|
| FARS | NHTSA | Every fatal traffic crash, 1975 onward | Deaths within 30 days only; latest year is preliminary |
| CRSS | NHTSA | Sample of police-reported crashes, 2016 onward | Estimates with error; not comparable to GES |
| NMVCCS | NHTSA | 2005 to 2007 crashes, light vehicles | A “critical reason” is not a cause or fault |
| LTCCS | FMCSA | 2001 to 2003 injury and fatal large-truck crashes | Associated factors are not causes; data over 20 years old |
| MCMIS | FMCSA | State-reported truck and bus crashes, inspections, carrier census | Preliminary for 22 months; depends on state reporting |
| SMS / CSA | FMCSA | Carrier percentiles built from MCMIS | Not a safety rating; property carrier percentiles withheld |
| State systems (e.g. TxDOT CRIS) | State DOTs | All reportable crashes in one state | State thresholds and vehicle definitions |
| IIHS Fatality Facts, NSC Injury Facts | IIHS; National Safety Council | Compilations and analyses | Definitions can differ from FARS |
FARS: the census of deaths
Publisher: NHTSA. What it counts: every fatal motor vehicle traffic crash in the 50 states, the District of Columbia and Puerto Rico since 1975. To qualify, a crash must involve a motor vehicle on a trafficway customarily open to the public and kill someone, in a vehicle or not, within 30 days12.
What it leaves out: deaths after day 30, crashes on private property not regularly used by the public2, and every crash in which nobody died.
Release lag: each year’s data arrive first as an Annual Report File (ARF), which is replaced the following year by a Final File. NHTSA’s large-truck fact sheet for 2024, built on the 2024 ARF, was published in June 2026. It also shows the scale of revision: the 2023 count of people killed in large-truck crashes moved from 5,472 in the ARF to 5,478 in the Final File2.
The trap: NHTSA’s FARS-based large-truck tables count any truck over 10,000 pounds gross vehicle weight rating, commercial or not, and exclude buses and motor homes2. A FARS “large truck” count and a “commercial vehicle” count from another source measure different fleets.
CRSS: the sample of everything else
Publisher: NHTSA. What it counts: a nationally representative sample of police-reported crashes of every severity, from property damage only to fatal, involving all types of motor vehicles, pedestrians and cyclists3. CRSS replaced the General Estimates System (GES) in 20162.
What it leaves out: crashes no police officer reported.
Release lag: NHTSA’s June 2026 large-truck fact sheet carried CRSS estimates through 20242.
The trap: CRSS numbers are estimates, not counts. FARS deaths are counts2. NHTSA warns that GES and CRSS estimates are not comparable because the sample designs differ2, so a trend line that runs from 2015 into 2016 splices two different instruments.
NMVCCS and LTCCS: one-time causation studies
Neither is an ongoing data series. Each sent investigators to crash scenes to reconstruct the events just before impact.
NMVCCS (NHTSA, National Motor Vehicle Crash Causation Survey) investigated crashes from July 3, 2005 to December 31, 2007. A weighted sample of 5,470 crashes represents about 2,189,000 crashes nationally4. To qualify, a crash had to occur between 6 a.m. and midnight, bring emergency medical services to the scene, and involve a light passenger vehicle that was towed5.
LTCCS (FMCSA, Large Truck Crash Causation Study) covered 967 crashes from 2001 to 2003 at 24 sites in 17 states, involving 1,127 large trucks, each crash with at least one death or injury. The weighted estimate is about 141,000 large trucks nationally7.
The trap: both studies assigned a “critical reason,” defined as the immediate reason for the pre-crash event that led to the first impact. NHTSA says it is not intended as the cause of the crash or as an assignment of fault4. A critical reason is the last link in a chain of events, not a verdict on who caused the crash. That caveat sits underneath the most-quoted figure in road safety: the driver was assigned the critical reason in 94 percent (±2.2%) of NMVCCS crashes4. In January 2022 NTSB Chair Jennifer Homendy urged NHTSA to stop using the figure, calling it misleading6. In LTCCS the truck was assigned the critical reason in 55 percent of crashes, and in two-vehicle crashes between a large truck and a passenger vehicle the passenger vehicle was assigned it in 56 percent and the truck in 44 percent7. The study’s “associated factors,” such as brake problems, coded for 29.4 percent of trucks, record what was present, not what caused the crash7. NHTSA also cautions against comparing NMVCCS estimates with other databases5. For how these figures are used, see what percentage of crashes are caused by human error and leading causes of large truck crashes.
A critical reason is the last link in a chain of events, not a verdict on who caused the crash.
MCMIS: the carrier file
Publisher: FMCSA, through its Analysis & Information (A&I) Online site. What it counts: crash, census and inspection files for commercial vehicles8. States report crashes involving a truck, a bus or a vehicle displaying a hazardous materials placard on a public roadway that result in a death, an injury or a towaway8. A&I’s large trucks and buses include commercial trucks over 10,000 pounds, vehicles designed to carry nine or more people, and placarded vehicles8.
What it leaves out: crashes states do not report, and crashes with no death, no injury treated away from the scene and no towaway.
Release lag: A&I considers MCMIS crash data preliminary for 22 months9. In October 2026 its home page still described 2025 crash numbers as provisional10.
The trap: an MCMIS crash record is a report that a carrier’s vehicle was involved, not a finding that the carrier was at fault.
SMS percentiles and what Congress withheld
The Safety Measurement System (SMS), part of FMCSA’s Compliance, Safety, Accountability program, scores carriers in seven safety categories and ranks them by percentile within each1112. FMCSA states that an SMS symbol is not a federal safety rating, and that a carrier is authorized to operate unless it has an Unsatisfactory rating or has been ordered to discontinue operations11.
Section 5223 of the FAST Act of 2015 barred public release of the relative percentile for each category, alerts, analysis of violations, and information on crashes in which the carrier or driver is found not at fault, until the Department’s Inspector General certifies that a list of conditions has been met, including a study of the program’s methodology and fixes for any deficiencies it found. It kept inspection and violation data, out-of-service rates and absolute measures public12. FMCSA states it has removed alerts and relative percentiles for property carriers from public SMS display13.
The trap: a property carrier’s category percentiles are not on the public SMS site. If a story cites one, ask where it came from and when.
State crash databases
Publisher: each state. TxDOT’s Crash Records Information System (CRIS), for one, holds reportable crashes from the Texas Peace Officer’s Crash Report, with summary reports published by June of the following year and a public query tool14.
What it counts: all reportable crashes under that state’s definition. In Texas that means injury, death or apparent damage of $1,000 to one person’s property15.
The trap: state vehicle categories do not match the federal ones. TxDOT counts a crash as a commercial motor vehicle crash when the officer marks the CMV box15. For 2024, TxDOT’s report counted 608 deaths in Texas CMV crashes16, while NHTSA’s FARS-based fact sheet counted 720 deaths in Texas crashes involving large trucks2. See Texas commercial truck crash statistics.
IIHS and NSC: compilations, not sources
IIHS Fatality Facts are annual summaries built on FARS, updated when DOT releases each year’s data17.
NSC Injury Facts counts motor-vehicle deaths differently. It includes traffic and non-traffic deaths, counts deaths within one year of the crash rather than 30 days, and uses death certificate data from the National Center for Health Statistics. For 2024 it reported 42,789 motor-vehicle deaths, set against NHTSA’s 39,254 traffic deaths18.
The trap: quoting an NSC total as a FARS number, or the reverse. Both are accurate under their own definitions.
