Crash Frequency Explains 3% of Whether You’ll Die

Two numbers define every vehicle in the FARS database: how often it shows up in a fatal crash, and how often its occupant dies when it does. The entire vehicle safety industry treats these as a single problem, multiplied together into a death rate and stamped on a press release.^[1] That multiplication hides something extraordinary.

Pearson r = 0.1825 across 272 vehicles with fleets exceeding 100,000 units and at least 50 fatal crash involvements. Squared, that gives us 3.3%, which means crash frequency predicts essentially nothing about crash lethality. A vehicle that crashes constantly might kill its occupants rarely, or always, and the frequency alone won’t tell you which.^[1]

This produces four distinct survival archetypes in the data, each requiring completely different engineering and policy responses.

Crashes Often, Kills Rarely

Ford’s E-350 Van appears in fatal crashes at 72.1 per 10,000 fleet vehicles, more than five times the median, yet its occupant lethality sits at just 41%. The Lincoln Navigator crashes at 31.3 per 10K with 42% lethality, the Toyota Tundra at 30.6 and 41%, Honda’s Odyssey at 25.8 and 43%.^[1] These are exposure machines with structural integrity. Heavy frames, commercial duty cycles, high annual mileage putting them in harm’s way constantly while their mass and crash structures keep occupants alive. The engineering works, but the usage pattern creates the risk.

Crashes Rarely, Kills Almost Always

Saturn’s S Series crashes at just 10.5 per 10K but kills at 92%, the Chevrolet Aveo manages 8.9 and 89%, Buick’s Park Avenue sits at 6.3 and 88%, and the Ford Escort barely registers at 2.5 frequency with 88% lethality.^[1] Low-mileage vehicles with aging, structurally compromised designs that keep their drivers out of trouble most of the time, but when physics finally arrives, there is nothing between the occupant and the dashboard except optimism. GM should have discontinued the S Series platform years before it actually did, and the Aveo was an embarrassment from the day Daewoo stamped the first body panel.

The Full Quadrant

Worst of both worlds: the Chevrolet S-10 at 86.4 frequency and 76% lethality, the Nissan Maxima at 88.8 and 66%, the Cobalt at 72.6 and 81%. Best of both: Kia’s Telluride at 1.4 frequency and 36% lethality with 31 total deaths across a 612,000-unit fleet, and Tesla’s Model 3 at 1.3 and 43% with 92 deaths across 1.575 million units.^[1] The spread from worst quadrant to best exceeds 60:1 on the composite, but the two dimensions generating that spread share almost no statistical relationship.

Class Averages Confirm the Split

Sports cars crash at 29.3 per 10K with 70.7% lethality, which makes them the worst-performing class on both axes simultaneously. Sedans follow at 21.6 and 65.4%, SUVs at 14.4 and 51.0%, and pickups crash at 26.8 with only 52.5%.^[1][2] Pickups crash nearly twice as often as SUVs but kill at comparable rates, because mass and frame geometry dominate lethality independent of how frequently the vehicle finds trouble. IIHS has documented this weight-survivability relationship for decades, but nobody has shown it operating independently of crash frequency across the full FARS dataset until now.^[3]

The Counterargument at Full Strength

This independence could be an artifact of confounders, not a mechanical truth. Old, cheap cars have high lethality because they are structurally weak AND low crash frequency because their elderly or low-income drivers accumulate fewer miles per year. New SUVs have low lethality because of modern engineering AND low frequency because suburban families drive predictable routes in predictable patterns. Vehicle age and owner demographics could independently drive both dimensions, making the low Pearson r a statistical artifact of third-variable confounding rather than proof that frequency and lethality are mechanically separate phenomena.^[2]

That argument deserves serious consideration. Controlling for vehicle age, driver age, and annual VMT at the model level would strengthen or destroy this finding, and FARS alone cannot do it. What FARS can show is that the traditional single-number death rate obscures which of two completely different failure modes is killing people in any given vehicle.

What This Means for Buyers

When shopping used, check both dimensions separately. A low overall death rate might mask high lethality hidden behind low exposure, the Saturn S Series problem: it barely crashes, but when it does, you are almost certainly dead. Conversely, a high death rate might reflect nothing more than commercial-grade mileage on a structurally excellent truck. The E-350 Van looks terrifying in FARS aggregate statistics, but its 41% occupant lethality is actually better than a Toyota Camry’s. Ask two questions, not one: how often does this model end up in FARS, and when it does, what percentage of occupants survive?^[1]

If you own anything in the high-lethality quadrant, check nhtsa.gov/recalls for outstanding safety campaigns. Structural weakness you cannot fix, but an unpatched airbag recall is a lethality multiplier you can eliminate in an afternoon.

Limitations

FARS captures only fatal crashes, roughly 36,000 annually out of an estimated 6.7 million total. Non-fatal crash frequency is unknown at the model level. Fleet size estimates use cumulative sales minus scrappage projections, introducing approximately ±15% uncertainty for low-volume models. Lethality includes all crash participants, not exclusively occupants of the subject vehicle. VMT estimates rely on class-level National Household Travel Survey data rather than model-specific odometer readings.^[4]