NHTSA Says Speeding Deaths Fell 11%. The 68x Gap Between Vehicles Didn’t Budge.

I ran the numbers twice, and they did not improve the second time. NHTSA just wrapped a month-long speeding campaign built on a real achievement: speeding-related fatalities fell 11% in 2025, from 11,288 to 10,035, the first return to pre-pandemic levels since 2019.^[1] Speeding accounts for 29% of all traffic deaths, and the agency spent July reminding America of exactly that.

But there is a number NHTSA did not mention, and it reframes the entire conversation. Take the entire FARS database (337 vehicle models, a decade of fatal crashes) and strip out every crash where the driver tested positive for alcohol or drugs, as though America solved impaired driving overnight. What remains is the sober death rate: fatalities per 100 million vehicle miles traveled among drivers who blew clean and tested clean.

Hyundai's Veloster, a discontinued hatchback lighter than most riding mowers cost, posts a sober death rate of 7·05 per 100 million vehicle miles, while Ram's 1500 sits at 0·10. That is a 68-to-1 gap, and not a single drop of alcohol explains any of it. In the Toyota Solara, where 96% of fatal crashes involved completely sober drivers and the impairment rate sits at just 4·1%, the sober death rate still reaches 4·08 per 100 million miles. Toyota's Land Cruiser, a vehicle synonymous with indestructibility in the countries that actually need off-road capability, posts a sober death rate of 5·71 with only 8·9% of its fatal crashes involving any substance at all.^[2]

Fleet-wide, impairment accounts for exactly 19·9% of the weighted-average fatality rate, which means four-fifths of the death rate persists even in a hypothetical America where nobody ever drives drunk or high. NHTSA's speed campaign targets 29% of deaths, so add those two behavioral categories together and campaigns, billboards, and enforcement can theoretically touch roughly half the problem. Everything else is structural: curb weight differentials that turn a 35-mph intersection into a physics execution, fleet age that keeps pre-ESC platforms circulating for decades after production stopped, and road geometries that funnel pedestrians into the path of vehicles engineered to protect only the people inside them.

That gap does not compress as you move down the ranking, either. Seven of the ten vehicles with the highest sober death rates are sedans or hatchbacks weighing under 3,500 pounds (the Cobalt at 3·96, the Impala at 3·93, the Maxima at 4·04), while nine of the ten safest are SUVs or trucks above 4,000 pounds, with the RAV4, Traverse, and Forester clustered at the bottom of the scale between 0·16 and 0·21. Strip impairment from the dataset entirely and what emerges is a weight map of America's fleet, drawn in bodies, where behavioral campaigns are succeeding at the margins while the structural gap between the vehicle a first-generation college student can afford and the one a suburban commuter leases has not shrunk by a single decimal point in ten years of data.

What you should do about it: If you are shopping for a vehicle, understand that your choice of car explains more of your crash survival odds than whether you drive sober, because a sober driver in a Veloster faces seven times the fatality risk of an impaired driver in a RAV4. Check IIHS ratings at iihs.org/ratings and compare vehicles in the same weight class rather than across them; if children ride in the back seat, look specifically for vehicles with a good rating in the updated moderate overlap front test, because many of the best-selling SUVs fail it.^[3] Run your VIN through NHTSA's recall database at nhtsa.gov/recalls while you are at it, because speed kills, and the vehicle you speed in determines whether you walk away.

Methodology note: “Sober death rate” is calculated by multiplying each vehicle’s FARS fatality rate (deaths per 100M VMT) by the fraction of its fatal crashes that did NOT involve alcohol- or drug-positive toxicology. This is a first-order approximation: it assumes impaired crashes and sober crashes share the same per-crash fatality weight, which they may not, and it cannot distinguish speed from other sober-crash causes (distraction, medical events, road design). The 68x gap is directionally robust but not precise to the decimal. FARS uses estimated VMT from NHTS annual mileage data, introducing ±15% uncertainty for low-volume models. See methodology for additional caveats.

Strongest counterargument: The sober death rate conflates vehicle design with driver demographics. The Veloster attracted young male drivers who speed more; the Ram 1500 is driven more on rural highways at steady speeds. Some of the 68x gap is selection bias in who buys what, not engineering. FARS cannot control for driver age, road type, or trip purpose. A randomized experiment (put identical drivers in a Veloster and a Ram and measure outcomes) would shrink the gap. By how much is unknowable from this data.