Impairment Doesn’t Explain Your Car’s Death Rate. We Did the Math.

Everyone has a theory about why certain vehicles pile up bodies in FARS data. Maybe the Mustang attracts reckless drunks. Maybe the Corvette is a midlife crisis on wheels, piloted home at 2 AM from a steakhouse bar. Reasonable assumptions. Also wrong. I cross-tabulated per-model FARS fatality rates with per-model toxicology results across 307 vehicle models and 191,193 fatal crashes, and the result guts the conventional wisdom: impairment is a near-constant across every vehicle class. It does not explain why some cars are 200 times deadlier than others per mile.

The Flattest Line in Crash Data

We grouped FARS toxicology results (BAC > 0 or drug-positive) by vehicle class:

Read that column again. Sports cars to minivans, the impairment rate spans 18.9% to 22.3%. A gap of three and a half percentage points. I ran the numbers twice because I didn't believe them. Meanwhile, the death rate per 100 million VMT ranges from 0.03 (Tesla Model Y) to 8.54 (Hyundai Veloster). A 285-to-1 ratio in outcomes. Explained by a factor that barely moves the needle.

The Sober Death Rate

I computed each vehicle's "sober death rate" by multiplying its per-100M-VMT rate by (1 minus its impairment fraction). Strip out every impaired driver. See what's left. What's left is the vehicle.^[1]

The Veloster drops from 8.54 to 7.05. Still a catastrophe. The Solara goes from 4.25 to 4.08, because only 4.1% of its fatal-crash drivers were impaired. Ninety-six percent of Solara occupants who died did so stone sober. Not a driver behavior problem. A sheet metal problem.

The Backward Correlation

If dangerous cars attracted impaired drivers, you would expect a positive correlation between death rate and impairment percentage. It runs backward. The highest-impairment vehicles are overwhelmingly performance and luxury cars with moderate death rates:

One in four Corvette drivers in fatal crashes tested positive. Yet the Corvette's death rate (1.52) is less than a third of the Veloster's (8.54), whose drivers are 9 percentage points more sober. The Corvette absorbs crashes. The Veloster does not. Good engineering covers for bad decisions. Bad engineering punishes good ones.

Methodology

For each of 307 models with 100+ drivers in FARS toxicology data (2014-2023), I multiplied the published per-100M-VMT death rate by (1 - anyPct/100), where anyPct is the percentage of drivers in fatal crashes who tested positive for alcohol (BAC > 0) or any drug. Output: an estimated sober death rate isolating vehicle-attributable risk from driver-behavior risk.^[2]

Limitations

This analysis assumes impairment is independent of crash outcome, which it is not. An impaired driver in a structurally weak car faces compounding risk. Our sober death rate is an approximation, not a causal decomposition. FARS toxicology testing is not universal; reporting rates vary by state, creating a 15-30% undercount of impaired drivers in some jurisdictions.^[3] Fleet age is a confound: older vehicles have both worse structural safety and different driver demographics. This impairment "constant" may reflect FARS sampling rather than true population behavior.

Strongest Counterargument

The uniformity of impairment rates across classes could be an artifact of FARS only capturing fatal crashes. Non-fatal crash data might show wildly different impairment distributions by vehicle type. If SUV occupants survive impaired crashes that sedan occupants do not, the fatality-level impairment rate converges artificially. This survivorship filter could mask real class-level behavioral differences that only appear at lower severity levels. NHTSA's CRSS (Crash Report Sampling System) data on non-fatal crashes would be needed to test this hypothesis, and we did not have access to model-level CRSS breakdowns.^[4]