FARS Says 9% of Fatal Crashes Involve Drugs. The Real Number Is Closer to 55%.
Forty-two thousand drivers in our FARS database tested positive for nonalcohol drugs. Out of 490,736 total drivers in fatal crashes between 2014 and 2023, that is 8.7%.[4] Alcohol positive: 15.1%. Total impaired: 20%. Those numbers feel manageable. Depressing, but manageable.
A peer-reviewed study published this month in Accident Analysis & Prevention analyzed 138,505 fatally injured drivers from the same FARS dataset and found something different.[1]
Between 2018 and 2022, after FARS expanded its toxicology data collection, 55.0% of fatally injured drivers tested positive for at least one nonalcohol drug. Up from 39.9% in 2013–2017.[1] Alcohol barely moved: 37.1% to 37.9%.
Drugs passed alcohol years ago. Nobody updated the scoreboard.
Three failures, one gap
Failure one: not everyone gets tested. On average, only about 60% of fatally injured drivers in the United States receive any drug testing at all.[3] That average masks state-level variation from above 90% down to below 10%. NHTSA, the GAO, and the NTSB have all flagged this inconsistency.[3] If your state barely tests, its contribution to the national tox dataset is mostly zeros — not because its drivers are sober, but because nobody checked.
Failure two: stop-limit testing. In many forensic toxicology labs, if a driver’s blood alcohol concentration hits 0.08% or above, the lab stops.[2] No drug panel. No further analysis. Budget constraints, “impairment already explained,” no enhanced penalties in most states. An NIJ-funded study of 2,514 suspected impaired-driving cases found 79% tested positive for drugs while only 40% tested positive for alcohol, and 23% had both.[2] Stop-limit testing catches the alcohol in those overlap cases and misses the rest entirely.
Failure three: FARS had a recording cap. Before 2018, FARS could store a maximum of three nonalcohol drugs per driver.[1] When that cap was lifted, the enhanced system increased the odds of detecting any nonalcohol drug by 42% (adjusted odds ratio: 1.42).[1] Drivers with three or more nonalcohol drugs in their system went from 2.0% to 5.8% of the sample — nearly tripling once there were enough fields on the form.[1] The drugs were always there. The database just didn’t have enough lines.
What’s actually in the blood
Marijuana positivity among fatally injured drivers: 26.4% in the 2018–2022 period, up from 14.4% before. Stimulants — primarily methamphetamine — rose from 9.3% to 25.1%, nearly tripling. Drivers positive for two or more nonalcohol drugs went from 13.6% to 19.2%. Drivers positive for both alcohol and marijuana: 8.3% to 11.8%.[1]
One in five fatally injured drivers had multiple nonalcohol drugs in their system. The poly-substance era arrived while the measurement infrastructure was still arguing about how to count single substances.
What this means for every number on this site
Every per-vehicle impairment statistic published here — every Charger bar-car percentage, every Corvette alcohol rate, every Prius sobriety badge — is a floor, not a ceiling. When we report that 4,339 Charger drivers show a 9.1% drug-positive rate, the real number is almost certainly multiples higher.[4] The same goes for every vehicle in the database. Our data captures what states chose to test, what labs chose to analyze after clearing the alcohol bar, and what FARS chose to record before it ran out of fields. It is a document of institutional attention, not pharmacological reality.
Limitations
The 55% figure from the published study reflects drivers who died in crashes, not all drivers involved. Our 8.7% aggregates across all drivers in fatal crashes, including surviving drivers who may not have been tested. These are different denominators, and comparing them directly overstates the ratio. The 6x gap is illustrative of the undercount problem, not a precise correction factor. Drug presence also does not equal drug impairment — a positive THC test can reflect use days earlier, and FARS makes no causal attribution between substance presence and crash causation.
Strongest counterargument
The jump from 39.9% to 55.0% may partly be a measurement artifact: FARS’s 2018 enhancement improved detection by 42%, meaning some of the apparent rise reflects better counting rather than more drugged driving. The study’s authors acknowledge this. But that reading actually strengthens the core finding. If lifting a three-drug recording cap instantly reveals 15 percentage points of previously invisible drug involvement, the old numbers were never measuring what we thought they were measuring. The real question was never “is drugged driving getting worse?” It was “how bad was the undercount all along?”