America’s Death Rate Hit a Historic Low. The Cars Doing the Killing Didn’t Get the Memo.

February 2026. NSC releases its preliminary estimate. 1.01 deaths per 100 million vehicle miles traveled.^[1] Lowest monthly figure in modern recordkeeping. Down from 1.32 just three years prior. A 23% decline. Headlines write themselves. America is getting safer behind the wheel.

Except America didn't do anything. The fleet changed underneath us.

I cross-referenced the NSC number against per-model death rates in FARS and found a split so clean it looks deliberate.^[2] Ninety vehicle models still carry death rates above 1.01 per 100M VMT. Those 90 models represent 36.2% of the registered fleet. They produce 131,090 of the 189,503 deaths in the dataset. That is 69.2% of all fatalities coming from barely a third of the cars on the road.

Meanwhile, 189 models sit below 1.01. Tesla Model Y at 0.03. Subaru Crosstrek at 0.08. Mazda CX-5 at 0.12. These vehicles are so safe in aggregate that they drag the national number down through sheer volume. Not because anyone learned to drive better. Not because distracted driving declined. Because enough people bought crossovers.

Four models run at five times the national rate or higher: Hyundai Veloster at 8.54, Chevy Tracker at 7.83, Toyota Land Cruiser at 6.27, Ford Mustang at 6.02.^[2] Twenty-eight models exceed double the national average, collectively responsible for 51,047 deaths from a fleet of just 14 million vehicles. And model years 2000 through 2008 account for 47.3% of all FARS fatalities despite their shrinking share of miles driven.^[2] Peak death model year: 2005, with 11,363 fatalities.

The national rate is an average. Averages are democratic. They count the Crosstrek and the Mustang equally. But the road doesn't.

NSC projects a 12% decline in total fatalities for full-year 2025.^[3] If that holds, it will be the largest single-year drop since seatbelt laws went national. And every bit of it will be compositional. Old sedans scrapping. Pre-ESC trucks aging out. MY2000-2008 vehicles finally disappearing from the denominator. The dangerous cars aren't getting safer. They're getting recycled.

What This Means for You

If you drive a vehicle from model years 2000 to 2008, your personal fatality risk may be three to five times the national rate regardless of how carefully you drive. Check your VIN at nhtsa.gov/recalls for outstanding recalls. If you are shopping for a used car and the budget puts you in a 2005 Mustang or a pre-2012 compact, know that the safety gap between that vehicle and a 2018 CX-5 is not incremental. It is categorical. Prioritize ESC, side curtain airbags, and a post-2012 model year. Those features are worth more than leather seats.

Methodology

FARS deaths (crash years 2014 to 2023) grouped by make and model. Per-model death rate = deaths / estimated fleet VMT (using NHTS average annual mileage scaled by estimated registrations). Models with rate above 1.01 per 100M VMT classified as "above national rate." 90 models above, 189 below. Fleet share = estimated registrations / total estimated registrations. Death share = model deaths / total FARS deaths (189,503). Model-year analysis sums deaths across all makes for each model year, 2000 to 2023.

Limitations

FARS covers 2014 to 2023 crash years and does not reflect 2024 or 2025 fleet changes. Per-model VMT estimates carry approximately 15% uncertainty, and low-volume models (under 50,000 fleet) can swing wildly on small death-count changes. The national rate (1.01) is from NSC preliminary data for February 2026 only, not a full-year figure, and is subject to revision. This analysis cannot separate fleet-composition effects from genuine per-vehicle safety improvements without year-by-year registration and VMT data by model, which FARS does not provide.

Strongest Counterargument

New vehicles ARE genuinely safer, not just compositionally dilutive. ESC, AEB, better crash structures, and improved roof strength mean a 2020 Mustang is meaningfully less lethal than a 2005 Mustang at the same speed. The fleet-composition framing risks understating real engineering progress. IIHS data confirms per-vehicle risk reductions of 30 to 50% for ESC-equipped models.^[4] This is fair. But the 90 models above 1.01 include post-2012 vehicles too. Composition explains the average. It does not excuse the outliers.