36,640: America's Record Low Is a Retirement Party, Not a Revolution

On April 1, NHTSA dropped a number that made highway safety professionals weep with cautious joy: 36,640 traffic deaths in 2025, down 6.7% from 2024. The fatality rate hit 1.10 per 100 million VMT. Second-lowest in recorded history. Every region of the country improved. Thirty-nine states got safer.

Beautiful. Genuinely. But before we declare victory over physics, it's worth asking a rude question: why is the number falling?

One answer lives in the FARS model-year data. Between 2014 and 2023, vehicles from model years 2003 through 2005 accumulated 33,298 deaths. That's the largest three-year death concentration in the dataset. MY 2004 alone is the peak death year for 17 different models, from the Chevrolet Cavalier to the GMC Envoy.^[1] These were pre-ESC, pre-backup-camera, pre-AEB vehicles built during the last great SUV and truck boom, sold cheap on used lots, and driven hard for two decades. In 2026, they're 21 to 23 years old. They're finally dying.

Not metaphorically. Literally scrapping out. And every Trailblazer that goes to the crusher takes its 43-deaths-per-year-of-exposure rate with it.

What's replacing the dead fleet?

If old vehicles dying were the whole story, pour the champagne. But FARS model-year data has a punchline: 13 vehicles have their highest annualized death rate in model year 2019, tied with 2004 as the most-concentrated peak year in the dataset.^[1] Add 2020 and 2021, and 30 vehicles peak in that three-year window. We're swapping one deadly generation for another.

Numbers worth sitting with: the Nissan Sentra kills at 34.6 deaths per year of exposure for MY 2019, up 72% from 20.1 for MY 2014. Kia Forte's MY 2021 rate is 25.3 per year. The Nissan Rogue climbed from 2.4 per year (MY 2010) to 21.4 (MY 2019).^[1] These aren't rust-bucket Cavaliers from a junkyard. These are current inventory on dealer lots.

Volume explains some of it. MY 2019 was a peak sales year, so more units means more road exposure. But volume is a multiplier, not a cause. A vehicle that sells 20% more units shouldn't produce 72% more deaths per year. Something in the driver population, the driving environment, or both shifted between 2014 and 2019, and the FARS data can't tell us which.

What still works

Don't let the existential dread overshadow the engineering. Three mandates are genuinely pulling people out of body bags. Electronic stability control, required since MY 2012, prevents an estimated 7,000 fatal crashes per year according to IIHS.^[2] Mandatory backup cameras arrived in MY 2018. AEB hit 90%+ of new vehicles by 2023.^[3] ESC alone may be the most effective safety regulation since seatbelt laws.

The catch: these are one-time gains baked into hardware. Once ESC saturation hits 100% of the active fleet, that savings curve flattens. We're within a few years of that ceiling.

The stubborn exceptions

Motorcycle deaths: up 1% in the first half of 2024 while every other road user category dropped.^[4] Truck-involved injuries: up 8.2% in 2024 even as truck-involved deaths fell 2.5%.^[5] Drivers over 65: up 1%.^[4] The falling headline number is averaging over pockets where things are getting worse, not better.

What this analysis can't prove

FARS covers 2014 through 2023. I'm inferring causes of a 2025 decline from decade-old fleet patterns. That's suggestive, not definitive. Model-year death peaks can't be perfectly separated from sales volume effects without registration-level fleet data that FARS doesn't include. The 2019-2021 death-rate spikes may partially reflect COVID-era driving behavior (empty freeways, triple-digit speeds) rather than vehicle-specific problems. And fleet attrition is a continuous process, not a sudden event; I can't isolate exactly how many of those 2003-2005 vehicles scrapped out between 2024 and 2025. NHTSA's detailed 2025 data, expected later this year, will resolve whether this improvement is structural or a statistical blip.

The math that should make you nervous

Model years 2003 through 2005 produced roughly 3,330 deaths per year during the FARS reporting window. Not all of those vehicles have scrapped out yet, but attrition accelerates sharply past age 15. If even half the remaining 2003-2005 fleet exited service between 2024 and 2025, the reduction in their death contribution would account for a significant chunk of the 2,614-death decline. Not all of it. But potentially most of it, without a single American changing their behavior behind the wheel.^[1]

36,640 is better than 39,254. That matters. But it's a retirement party for the 2003 Trailblazer, not a reckoning with the 2019 Sentra. And retirement parties end.

What you can do

Check your vehicle's model year against FARS trends at NHTSA's query tool. If you're shopping for a used car, avoid the 2003-2005 vintage entirely; they're cheap for a reason. Prioritize ESC (standard since 2012) and AEB (common since 2018) when buying used. And if you ride a motorcycle, understand that the national safety improvement does not include you.