36,640 Dead. The Decline Is Real. Half of It Is Just Old Cars Dying.

I ran the numbers, then I ran them again, and they still told me the same thing: NHTSA announced on April 1 that 36,640 people died on American roads in 2025, a 6.7% drop from 2024's 39,254.^[1] Fifteen consecutive quarters of decline, a fatality rate of 1.10 per 100 million VMT, the second-lowest ever recorded and trailing only 2014's 1.08. Every outlet ran the headline, but nobody ran the decomposition.

So I did. Our FARS dataset tracks 337 models across 187,058 deaths from 2014 to 2023, broken down by model year of the vehicle involved.^[2] Of those fatalities, 134,491 involved vehicles built before model year 2012, the year electronic stability control became mandatory under FMVSS 126.^[3] That is 71.9% of all deaths concentrated in vehicles that predate a technology proven to cut fatal single-vehicle crashes by 56% and multi-vehicle fatal crashes by 32%.^[4]

The deadliest model year cohort? 2004 through 2007, accounting for 42,981 deaths in a decade of FARS data. Those vehicles are now 19 to 22 years old, deep into the scrappage curve's steepest gradient. S&P Global Mobility pegs the average vehicle on U.S. roads at 12.6 years, which means the median car was built right around the ESC mandate cutoff.^[5] Every year, roughly 6–8% of the surviving pre-ESC fleet gets crushed, recycled, or abandoned in a field. Replaced, eventually, by vehicles that won't kill you in a routine overcorrection at 45 mph.

Rough math: if 7% of the pre-ESC fleet scraps annually, and pre-ESC vehicles die at approximately 2.5 times the rate of post-ESC vehicles per mile traveled, fleet turnover alone generates a 3–4% annual fatality decline. The actual 2025 decline was 6.7%, which puts the structural contribution from fleet replacement somewhere around half the total improvement, with the remainder attributable to AEB adoption, road design upgrades, and whatever behavioral normalization is happening as post-COVID recklessness fades.

Fair counterpoint: pedestrian deaths also fell 11% in the first half of 2025, per GHSA tracking.^[6] ESC does nothing for pedestrians, and neither does fleet age, and the federal AEB pedestrian-detection mandate doesn't kick in until 2029. Something else is pulling those numbers down, and I genuinely cannot attribute it to fleet mechanics. Behavioral normalization, improved crosswalk infrastructure, or statistical noise in a one-year sample? Pick your favorite; the data doesn't resolve it yet.

What the data does resolve: the U.S. still kills people at 2.4 times the rate of Western European peers. Sweden and Norway sit at 0.20 per 100M VMT,^[7] and we celebrated reaching 1.10. That gap is not closing through fleet turnover. It requires speed management, road geometry, and political will that no scrappage curve can substitute for.

What this means for you: If you drive a pre-2012 vehicle, your car lacks the single most impactful crash-prevention technology ever mandated. ESC is not a luxury feature; it is the difference between a controlled recovery and a rollover at highway speed. Check your VIN at nhtsa.gov/recalls for open recalls, and when you replace that car, prioritize models with IIHS Top Safety Pick ratings from 2018 or later.