Understanding traffic safety data: how to read crash statistics wisely
How crash numbers are counted, what they miss and how to compare rates fairly
Picture a local news story saying traffic deaths in your county jumped 50% last year. It sounds alarming, but if deaths rose from 4 to 6, chance alone could explain much of the change.
Crash numbers shape real choices, from where a city paints a crosswalk to which laws a state passes. Epidemiologists judge each number by asking how it was made, what it leaves out and what it is being compared with.
This guide explains where U.S. crash data come from, which numbers are solid, how rates work and how to spot a misleading comparison.
Where U.S. crash numbers come from
Most crash data begin with a police report, written on a form that each state designs. A voluntary federal guideline called the Model Minimum Uniform Crash Criteria (MMUCC), now in its 2024 edition, sets out a common minimum list of data fields [4]. Even so, the National Highway Traffic Safety Administration (NHTSA) notes that definitions and fields still vary from state to state [4].
FARS counts every fatal crash. NHTSA’s Fatality Analysis Reporting System (FARS) is a census rather than a sample, with yearly data back to 1975 [1]. A crash qualifies if it involves a motor vehicle on a road open to public travel and someone dies within 30 days [1]. Trained state employees called FARS analysts pull facts from police reports, death certificates, medical examiner and EMS reports, and driver and vehicle records [1]. They code well over 100 standard details for each crash. Software flags entries that don’t add up [1].
CRSS samples everything else. The Crash Report Sampling System (CRSS) is a nationally representative sample of police reports, used to estimate injury and property damage crashes nationwide [2]. It replaced the older General Estimates System in 2016. NHTSA warns that the two are not comparable [2]. Because CRSS is a sample, its estimates carry error margins. NHTSA flags some changes as not statistically significant, like 2024’s 0.8% drop in injuries [2].
Injury severity is a roadside judgment. Officers rate each person’s injury on the KABCO scale [4]. K means fatal, A suspected serious injury, B suspected minor injury, C possible injury and O no apparent injury [4]. The word “suspected” matters, because an officer’s call at the scene is not a medical diagnosis.
Other lenses. Health agencies count deaths from death certificates [14]. Insurers see many crashes that police never hear about. The Highway Loss Data Institute (HLDI), an affiliate of the Insurance Institute for Highway Safety (IIHS), studies insurance claims [15]. Its data come from companies representing over 85% of the U.S. private passenger vehicle insurance market [15]. Each system follows its own rules, so the totals differ.
Comparison compiled by Safety Behind the Wheel Foundation
| 2024 measure | Who counts | What it includes | Figure |
|---|---|---|---|
| Traffic deaths | NHTSA (FARS) | Deaths within 30 days of a crash on a public road | 39,254 [2] |
| Motor vehicle traffic deaths | CDC (death certificates) | Deaths certified as motor vehicle traffic deaths | 41,241 [14] |
| Motor vehicle deaths | National Safety Council (NSC) | Traffic and nontraffic deaths within 1 year of the crash | 42,789 [13] |
| People injured | NHTSA (CRSS) | People hurt in police reported crashes | 2.42 million [2] |
| Medically consulted injuries | NSC | NSC’s estimate of injuries that involved medical care | 4.9 million [13] |
Each answers a different question, so never mix them in one comparison.
Early estimates, annual files and final counts
Each year’s death count arrives in three stages, each more complete than the last:
- Early estimate, the next spring. NHTSA models the count from early state reports and other monthly tallies. In April 2026, it estimated 36,640 deaths for 2025, down 6.7% [3].
- Annual Report File, about a year later. This first full FARS file is the one NHTSA’s April 2026 overview of 2024 uses [1, 2].
- Final File, about a year after that. It adds late cases and updates. NHTSA expects the 2024 final file in early 2027 [1, 3].
Revisions are usually small: the 2023 count rose from 40,901 in the annual file to 41,025 in the final file, a 0.3% change (our calculation) [2]. Even small revisions matter when a headline hinges on a small change.
Fatal crashes are counted best; injury crashes are not
A death is hard to miss. FARS draws on death certificates as well as police reports [1]. That makes fatal counts the most complete figures in traffic safety.
Injury and property damage crashes are a different story, because many never reach police. NHTSA’s study of the economic and societal impact of crashes, using 2019 data, estimated [5]:
- About 14.2 million crashes in all, of which about 6.6 million were police reported.
- About 4.5 million people injured, with 32% of those injuries in crashes police never recorded.
- About 60% of property damage only crashes and 32% of injury crashes went unreported.
The same study counted 36,500 deaths in 2019, including 145 people who died more than 30 days after the crash and so fall outside FARS [5]. The lesson is to treat injury and damage totals as floors. When they change, ask whether reporting habits changed too.
Why distraction and drowsiness are undercounted
Alcohol can be measured in blood or breath, but there is no roadside test for distraction or drowsiness. An officer needs clues, a witness or an admission. A driver who died can’t explain.
NHTSA spells out the limits for distraction [6]:
- Crash report forms differ. Not all capture distraction the same way.
- Drivers may not admit to texting or other risky behavior.
- After a fatal crash, police may have no way to learn what the driver was doing.
- Forms are slow to catch up with new devices.
Within those limits, police data tied 3,208 deaths in 2024 to distraction, about 8% of the total [2].
For drowsiness, NHTSA says a precise count isn’t yet possible and that experts broadly agree its figures understate the problem [7]. Police data linked 644 deaths in 2024 to drowsy driving, 1.6% of the total [2]. Two other lenses point to a much larger role:
- Statistical modeling. Using in depth crash investigations to fill gaps, the AAA Foundation for Traffic Safety estimated that 17.6% of fatal crashes in 2017 to 2021 involved a drowsy driver [8]. Those crashes killed an estimated 29,834 people, about 6,000 a year (our calculation) [8].
- Naturalistic driving. The second Strategic Highway Research Program (SHRP2) recorded 3,593 volunteer drivers on video from 2010 to 2013 [9]. Drowsiness showed up in 8.8% to 9.5% of their crashes, and in 10.6% to 10.8% of crashes serious enough to report to police [9].
Alcohol is measured better, but not perfectly. When a driver’s test result is missing from FARS, NHTSA estimates it with a statistical method called multiple imputation [10]. A crash counts as alcohol impaired when a driver had a blood alcohol concentration (BAC) of .08 g/dL or higher [10]. NHTSA notes that this label shows an impaired driver was involved, not that alcohol caused the crash [10]. Our guides to alcohol impaired driving and drowsy driving cover prevention.
Rates: choosing the right denominator
A count tells you how much harm happened, while a rate tells you how risky something is. A rate divides harm by exposure, meaning how much people do the thing that can lead to a crash. Epidemiologists call that bottom number the denominator. Choosing it is most of the job.
Comparison compiled by Safety Behind the Wheel Foundation
Which number answers my question?
| Your question | Best measure | 2024 national figure | Watch out for |
|---|---|---|---|
| Where is the most harm, so where should help go first? | Count of deaths or serious injuries | 39,254 deaths [2] | Big places nearly always have bigger counts |
| How risky is each mile of travel? | Deaths per 100 million vehicle miles traveled (VMT) | 1.19 [2] | Miles driven by visitors count where they drive |
| How heavy is the toll on a community? | Deaths per 100,000 people | 11.5 [11] | Deaths count where the crash happened, even for visitors [1] |
| Is one group of drivers riskier than another? | Crashes per licensed driver, and per mile | Varies by group | People who drive little can look safe per driver |
| Is one kind of trip riskier than another? | Crashes per trip or per hour of travel | Rarely published | Depends on travel surveys, which have their own errors |
Local risk versus national risk
Rural and urban. In 2024, rural areas had 21% of Americans and 31% of vehicle miles, but 41% of traffic deaths [12]. Their death rate of 1.56 per 100 million miles was 1.5 times the urban rate of 1.01 [12]. NHTSA classifies each crash by where it happened, using urban boundaries set by states and approved by the Federal Highway Administration [12]. Our rural road safety guide explains the gap.
State differences. IIHS reports that in 2024, deaths per 100,000 people ranged from 4.7 in Rhode Island to 25.6 in Mississippi [11]. Per mile, rates ran from 0.59 in Massachusetts to 1.81 in Mississippi [11].
Small numbers swing. In a town with 4 deaths one year, 2 more makes a 50% jump. NHTSA’s early state estimates show the effect even for whole jurisdictions [3]. The District of Columbia fell 46.8%, from 47 deaths in 2024 to an estimated 25 in 2025. Hawaii rose 26.5%, from 102 to 129 [3]. Swings that large in small places often reflect chance as much as real change. So look at 3 to 5 years together.
Regression to the mean. Statisticians use this term for a predictable pattern: a place with an unusually bad year tends to have a calmer one next, even if nothing changes. If a road gets a fix right after a spike, part of the next year’s drop may be chance. That is why sound before and after studies adjust for it. To explore your own area, see our guide to local crash data and safety maps.
Worked example: one state, two stories
The same 2024 data can support opposite headlines [11]:
- Story A: Wyoming’s roads are among the deadliest. Wyoming had 18.2 deaths per 100,000 people, 58% above the national 11.5 and sixth highest among the 50 states and DC.
- Story B: Wyoming’s roads are safer than average. Wyoming had 1.10 deaths per 100 million miles, about 8% below the national 1.19.
Both stories rest on the same 107 deaths. Only the denominator changed. Wyoming’s roads carried about 16,500 vehicle miles per resident, compared with about 9,700 nationwide (our calculation from IIHS figures) [11]. Long rural distances and through travel can pile up miles on a small population.
The District of Columbia flips the pattern, with 6.7 deaths per 100,000 people, among the lowest, but 1.33 per 100 million miles, above average [11]. Its roads carried only about 5,000 miles per resident (our calculation). So the toll per person stays low even though each mile carries more risk.
Both stories are right, but they answer different questions. To judge how safe each trip is, lean on the per mile rate. To size the burden on residents, use the per person rate. A responsible headline names its measure.
Misleading comparisons to avoid
- Raw counts treated as risk. Texas recorded 4,160 deaths in 2024 and Wyoming 107, yet Wyoming’s rate per person was higher [11].
- Percent change from a small base. A 50% rise can mean 2 more deaths, so ask for the counts behind any percentage.
- “Most crashes happen near home.” Even when true, this mostly reflects where people drive. Most trips start or end near home, so the claim doesn’t prove your neighborhood is riskier per mile.
- Alcohol involved versus alcohol caused. An alcohol impaired crash had an impaired driver involved, which alone doesn’t prove cause [10]. Labels like speeding related, which covered 29% of 2024 deaths, work the same way [2]. See our speeding guide.
- Correlation versus causation. If deaths fell the year a new law took effect, the law may deserve credit, or travel, weather or the economy may. Strong studies compare against places or periods without the change.
- Comparing across definition changes. CRSS can’t be compared with the older General Estimates System. NHTSA’s vehicle type categories since 2020 don’t match 2019 and earlier [2]. FARS also changed how it records driver distraction in 2010 [1].
- Mixing crashes, vehicles and people. In 2024, 36,297 fatal crashes killed 39,254 people and involved about 55,600 vehicles [13]. Fatal crashes fell 3.9% while deaths fell 4.3%, which are two different facts [2].
- Early estimates versus final counts. Compare early with early, and final with final.
Frequently asked questions
How many people die in car crashes in the U.S. each year?
NHTSA counted 39,254 traffic deaths in 2024 and estimates 36,640 for 2025 [2, 3]. CDC and the National Safety Council report higher 2024 totals because they count deaths under different rules [13, 14].
What is the difference between FARS and CRSS?
FARS is a census of crashes on public roads in which someone dies within 30 days [1]. CRSS is a nationally representative sample of police reported crashes, used to estimate injuries and damage only crashes [2].
Are distracted and drowsy driving statistics accurate?
Read them as minimums, since NHTSA acknowledges that police often can’t detect distraction or drowsiness [6, 7]. Video studies and statistical models suggest drowsiness plays a role in far more crashes than police data show [8, 9].
Which state has the deadliest roads?
That depends on the measure, though in 2024 Mississippi had the highest death rate both per person and per mile [11]. Many other states shift rank with the measure, as Wyoming shows.
Why does final crash data take so long?
States keep gathering death certificates, medical examiner reports and late updates after the year ends. NHTSA releases an annual file after the crash year and replaces it with a final file about a year later [1].
The bottom line
Recommendation from Safety Behind the Wheel Foundation
Before you trust or share a traffic number, ask who counted it, what it leaves out and what it is compared with. Use counts to size a problem and rates to judge risk. Treat any single year, small place or early estimate as a clue rather than a verdict.
Related articles
- Local crash data and safety maps: how to find and read your area’s data
- Rural road safety: why country roads are deadlier and how to drive them
- Drowsy driving: warning signs, hidden risks and what really works
- Alcohol impaired driving: know the risks and plan a sober ride home
- Speeding and speed management: why a few mph matter so much
Sources and further reading
- 1National Highway Traffic Safety Administration (NHTSA). Fatality Analysis Reporting System (FARS) web page, and Fatality Analysis Reporting System Analytical User’s Manual, 1975 to 2023. DOT HS 813 706. April 2025. nhtsa.gov/…/fatality-analysis-reporting-system-fars and crashstats.nhtsa.dot.gov/…/813706
- 2National Highway Traffic Safety Administration (NHTSA). Overview of Motor Vehicle Traffic Crashes in 2024. Traffic Safety Facts Research Note, DOT HS 813 791. April 2026. crashstats.nhtsa.dot.gov/…/813791
- 3National Highway Traffic Safety Administration (NHTSA). Early Estimate of Motor Vehicle Traffic Fatalities and Fatality Rate in 2025. Traffic Safety Facts Crash•Stats Brief Statistical Summary, DOT HS 813 800. April 2026. crashstats.nhtsa.dot.gov/…/813800
- 4National Highway Traffic Safety Administration (NHTSA). MMUCC web page, and MMUCC Guideline: Model Minimum Uniform Crash Criteria, 6th Edition. DOT HS 813 525. Revised February 2025. nhtsa.gov/mmucc-1 and crashstats.nhtsa.dot.gov/…/813525
- 5Blincoe L, Miller T, Wang J-S, et al. The Economic and Societal Impact of Motor Vehicle Crashes, 2019 (Revised). NHTSA, DOT HS 813 403. February 2023. crashstats.nhtsa.dot.gov/…/813403
- 6National Highway Traffic Safety Administration (NHTSA). Distracted Driving in 2024. Research Note, DOT HS 813 790. April 2026. crashstats.nhtsa.dot.gov/…/813790
- 7National Highway Traffic Safety Administration (NHTSA). Drowsy Driving. Accessed October 2026. nhtsa.gov/risky-driving/drowsy-driving
- 8Tefft BC. Drowsy Driving in Fatal Crashes, United States, 2017 to 2021. AAA Foundation for Traffic Safety. March 2024. aaafoundation.org/drowsy-driving-in-fatal-crashes-united-states-2017-2021
- 9Owens JM, Dingus TA, Guo F, et al. Prevalence of Drowsy Driving Crashes: Estimates From a Large-Scale Naturalistic Driving Study. AAA Foundation for Traffic Safety. February 2018. aaafoundation.org/prevalence-drowsy-driving-crashes-…
- 10National Highway Traffic Safety Administration (NHTSA). Alcohol-Impaired Driving: 2024 Data. Traffic Safety Facts, DOT HS 813 815. June 2026. crashstats.nhtsa.dot.gov/…/813815
- 11Insurance Institute for Highway Safety (IIHS). Fatality Facts 2024: State by State. Posted June 2026. iihs.org/…/state-by-state
- 12National Highway Traffic Safety Administration (NHTSA). Rural/Urban Traffic Fatalities: 2024 Data. Traffic Safety Facts, DOT HS 813 828. July 2026. crashstats.nhtsa.dot.gov/…/813828
- 13National Safety Council (NSC). Injury Facts: Motor Vehicle Overview, Introduction. Last modified April 16, 2026. injuryfacts.nsc.org/motor-vehicle/overview/introduction
- 14Centers for Disease Control and Prevention (CDC), National Center for Health Statistics. FastStats: Accidents or Unintentional Injuries (National Vital Statistics System mortality data, 2024). Last reviewed February 18, 2026. cdc.gov/nchs/fastats/accidental-injury.htm
- 15Insurance Institute for Highway Safety (IIHS). About Us (Highway Loss Data Institute). Accessed October 2026. iihs.org/about-us
This article reflects the priorities of traffic safety statistics and epidemiology. It is general education, not legal, medical or professional advice for any individual. Crash data, definitions and methods change. So check the latest releases from NHTSA, CDC, IIHS and your state highway safety office. Reviewed October 2026.
Keep learning
Safe Driving Resource Center
Guides, self-checks and articles for drivers, parents of new drivers, and families recovering after a crash.
