NHTSA's Crash Database Undercounts Truck Underride Deaths by 46%. The Broken Number Sinks the Fix.

In 2016, a Tesla Model S drove under the side of a semitrailer in Williston, Florida, killing Joshua Brown instantly. NTSB investigators documented every detail, and Brown's death became the most scrutinized underride fatality in American history. And when researcher Marianne Karth looked it up in NHTSA's Fatality Analysis Reporting System, the database said: "No underride or override noted."

That is not a glitch. Independent researcher Eric Hein examined 40 verified fatal side underride crashes and cross-referenced each one against the FARS record.^[1] Of 39 crashes found in the database, 16 were coded as "No Underride or Override Noted." One crash that killed three people was missing entirely. Total damage: 17 of 40 crashes and 28 of 61 fatalities either miscoded or absent. ProPublica ran a similar check: 27 fatal underrides confirmed in one NHTSA database, three coded correctly in FARS.^[2] Three out of twenty-seven.

This matters because NHTSA does not mandate safety equipment on vibes — it runs cost-benefit analysis. The agency calculated that requiring side underride guards on every new semitrailer would cost between $973 million and $1.2 billion and would prevent exactly 17.2 deaths per year.^[3] After applying discount rates and amortizing costs over the full fleet replacement cycle, NHTSA's own ANPRM put the cost per equivalent life saved between $73.5 million and $103.7 million. The agency's regulatory threshold sits around $12.5 million. Side guards fail by a factor of six.

But IIHS Senior Research Engineer Matthew Brumbelow reran the math with broader crash inclusion criteria, event data recorder speeds instead of posted limits, and adjustments for multi-vehicle pileups NHTSA excluded.^[4] His estimate: 159 to 217 passenger vehicle occupant deaths preventable per year. Run those numbers through the same discounted cost framework and the per-life figure drops to roughly $4.5 to $7.5 million — below NHTSA's threshold, and the mandate passes. Europe has required lateral protection devices on heavy trucks since the 1980s under ECE Regulation 73. Four decades of real-world data from countries that actually mandated the fix, and NHTSA still treats the question as open.

Seventeen states have a field on their police crash reports for underride. Thirty-three do not.^[2] Officers in those states literally cannot record what killed the occupants. The data gap starts at the scene, propagates through state reporting, and arrives at FARS already sanitized. NHTSA acknowledged in its 2022 Advance Notice of Proposed Rulemaking that its underride data is unreliable, then used that data anyway to conclude the fix costs too much.^[3]

Congress has introduced the Stop Underrides Act four times since 2017. Filed as S. 3775 and H.R. 7354, the 2026 version includes a provision requiring NHTSA to review FARS and correct misclassified underride crashes.^[5] Read that again: Congress is writing legislation to tell the safety agency to fix its own database. It also mandates free law enforcement training on identifying underride crashes at scenes, tacitly admitting the data pipeline has been broken at every stage for decades.

The trucking industry's counterarguments deserve full weight, and there are several. Advanced driver assistance systems may reduce underride crashes independently as fleet penetration climbs. Guard weight reduces cargo capacity on every load. Small carriers operating on single-digit margins face a disproportionate cost burden. And proposed guard designs have been tested at impact speeds up to 40 mph, while many fatal underrides occur at highway speed, where structural adequacy is less certain. If ADAS cuts the target population by half over the next decade, even IIHS's higher estimate might not justify a billion-dollar mandate, and that uncertainty is real. But the people dying under trailers right now will not benefit from automatic braking systems installed on vehicles manufactured in 2034. And ADAS fleet penetration will not reach saturation for twenty years at current replacement rates.

Because FARS only records fatal crashes, non-fatal underride collisions produce traumatic brain injuries, spinal cord damage, and amputations at rates nobody has systematically counted. Hein's sample of 40 crashes is small; the true FARS miscoding rate could be higher or lower. IIHS and NHTSA used different crash inclusion criteria, and neither methodology has been independently validated by a third party.

What to do with this: On the highway, don't linger alongside a semitrailer. Pass decisively or drop back. Trucks have no side guards, no side-impact airbags, and a ground clearance that invites a sedan's roof to become the first point of contact. When buying a vehicle, prioritize models with top IIHS crashworthiness ratings and advanced automatic emergency braking, which can reduce the closing speed that makes underride fatal. The Stop Underrides Act 2.0 (S. 3775 / H.R. 7354) is currently before committee, and NHTSA's public comment docket for the side underride guard ANPRM (RIN 2127-AM54) remains open at regulations.gov. The database error that sinks the mandate is documented and fixable. Fixing the count fixes the math. Fixing the math saves between 159 and 217 people a year.