Impairment Is a Constant. The Vehicle Is the Variable.

According to the toxicology reports, and there are a lot of them, the Pearson correlation between a vehicle’s death rate and its driver impairment rate across 200 models in FARS is -0.195.^[1] Negative. The cars that kill the most people per mile driven have lower rates of impaired drivers than the cars that kill the fewest. I ran it twice. The sign did not change.

NHTSA launched its “Drive Sober or Get Pulled Over” campaign in July with $665 million in safety grants, celebrating that impairment-related fatalities dropped 11% in 2025.^[2] Fair enough. But the FARS data says something the campaign materials do not: impairment is not what separates a safe car from a coffin on wheels. The vehicle is.

Split 200 vehicles into quartiles by death rate. The safest 25% average a rate of 0.33 deaths per 100 million VMT. Their average impairment rate: 19.5%. The deadliest 25% average a rate of 2.92, nearly nine times higher. Their average impairment: 19.6%.^[1] A rounding error separates them. The death rate multiplies by nine. The impairment stays flat.

The deadliest quartile accounts for 86,103 of the 184,090 deaths in the dataset. Nearly half the body count. And their drivers are no more impaired than the drivers of a Toyota RAV4 or a Honda CR-V sitting in the safest quartile. A Hyundai Veloster posts a death rate of 8.54 with 17.4% impairment. A Chevy Tracker hits 7.83 with 12.7%. The Toyota Solara reaches 4.25 with 4.1% impairment, meaning ninety-six percent of its fatal crash drivers were stone sober.^[1] These vehicles are not killing impaired drivers. They are killing everyone.

Forty-six of the 200 vehicles sit in what I will diplomatically call the engineering-failure quadrant: above the median death rate and below the median impairment rate. Twenty-three percent of the fleet occupies the space where the car is deadly and the driver was doing nothing detectably wrong. No amount of sobriety checkpoints will fix a crumple zone that was never designed to absorb the energy of the crashes these vehicles actually experience.

Strongest counterargument: the correlation is weak, and impairment measurement in FARS has known limitations. Toxicology testing rates vary by state and jurisdiction. Some states test every fatally injured driver; others test fewer than half. A vehicle popular in a low-testing state might show artificially low impairment simply because nobody checked. Additionally, the death rate in FARS reflects the vehicle population over the 2014 to 2023 window, so high-rate vehicles tend to be older, discontinued models with smaller fleets and more structural risk, which correlates with different demographics independently of impairment. The negative correlation does not prove that sobriety causes crashes; it shows that impairment does not predict which vehicles kill the most, and that distinction matters for policy.

What this means for you: If you are shopping for a car and a salesperson tells you safety is mostly about driver behavior, ask them to explain why a Hyundai Palisade at 0.06 and a Hyundai Veloster at 8.54 share a manufacturer logo but have a 142-fold gap in death rate at nearly identical impairment levels. Check the FARS data for your specific model at the Crash Report. Then check the IIHS crashworthiness rating. A sober driver in a dangerous car is still in a dangerous car.