Some Brands Are Safe Bets. Others Are Russian Roulette.

Walk into a Subaru dealership, close your eyes, and point. Whatever you land on kills at a rate below 1.0 per 100 million VMT.^[1] Outback: 0.45. Forester: 0.26. Legacy: 0.95. Impreza: 0.52. Ascent: 0.78. Every single model, below the national sedan average. Standard deviation across the lineup: 0.24. You literally cannot pick wrong.

Try that at Chevrolet. The Traverse posts 0.2 deaths per 100M VMT—one of the safest vehicles in FARS. The Tracker: 7.83. Not a rounding error. A 39-fold gap in fatality rate, wearing the same bowtie.^[1] Twenty-five qualifying models, standard deviation 1.81. The badge told you nothing.

I calculated the within-brand standard deviation for every manufacturer with four or more models recording 200+ deaths in FARS 2014–2023.^[1] Death rate per 100M VMT, using NHTS fleet-weighted mileage estimates.^[3] Low σ = consistency. High σ = roulette.

Kia matched Subaru. Six models, 0.28 to 1.07, σ = 0.27.^[2] Forte: 0.4. Optima: 0.58. Soul: 0.64. Not spectacular. Just consistently refusing to kill you.

Hyundai—Kia’s corporate sibling, same platform engineering group—scored the worst σ of any brand: 2.88. Tucson at 0.34, Veloster at 8.54. Same parent company, 25-fold safety gap.^[1] Whatever consistency Kia inherited from that partnership, Hyundai did not.

Toyota should worry more people than it does. The RAV4 records a 0.19 death rate. The Land Cruiser: 6.27. Thirty-three times deadlier.^[1] Fourteen models, σ = 1.62. “I buy Toyotas because they’re safe” is not a complete sentence. Which Toyota?

Now the rebuttal, because it’s a real one: Chevrolet makes everything from subcompact sedans to body-on-frame trucks to a mid-engine Corvette. The Tracker was discontinued in 2004. Comparing it to a 2020 Traverse measures two eras and two vehicle classes under one logo. Subaru’s tight σ partially reflects a narrow, homogeneous lineup rather than superior engineering.

Fair. Also irrelevant to the buyer. If you walked into a Chevy dealer in 2018 because “Chevrolets are solid,” the brand name told you exactly zero about whether your specific model would protect you. FARS doesn’t care about your brand loyalty. It counts bodies.

Caveats: FARS captures fatal crashes only—not the 6.7 million annual crashes that don’t kill anyone.^[4] VMT estimates carry ±15% uncertainty on low-volume models. Driver demographics vary wildly between a Veloster and a Tucson. We’re measuring the vehicle-driver-road system, not the vehicle alone. And occupant death rates ignore what these vehicles do to the cars they hit.

But brand is not safety. Model is safety. If your research starts and ends at a logo, you’re bringing a marketing department to a physics exam.