America’s Death Rate Is 1.10. The Cars Say It Should Be 0.75.

NHTSA released its 2025 preliminary estimates three weeks ago, and the number everyone fixated on was 1.10 fatalities per 100 million vehicle miles traveled.^[6] Second-lowest in recorded history, behind only 2014’s 1.08. Celebrations ensued. The administrator called American roads “safer.” Thirty-nine states posted declines.^[7]

Nobody asked the obvious question: why is the rate almost identical to 2014’s, when the cars are radically different?

Consider what changed between those two fleets: in 2014, roughly 40% of vehicles on U.S. roads had electronic stability control, and by 2025 that number exceeds 85%. IIHS found ESC cuts fatal single-vehicle crash risk by 49% for SUVs and 33% for passenger cars.^[1] Side-curtain airbags, rare on budget vehicles a decade ago, are now standard across virtually every nameplate. IIHS documented a 37% reduction in fatal near-side crash risk from curtain airbags alone.^[2] Automatic emergency braking, nonexistent in 2014, covers about half the new vehicles sold since 2020 and cuts front-to-rear crashes by roughly 50%.^[3]

Then there is the crossover migration. In 2014, sedans and coupes accounted for approximately 45% of new-vehicle sales; today they are below 20%. That matters because FARS data covering 183,188 fatalities from 2014 to 2023 shows a relentless pattern: sedans have a 35.7% occupant survival rate in fatal crashes, while SUVs manage 47.4% and pickups 51.2%.^[4] Every buyer who swapped a Malibu for an Equinox didn’t just get a taller ride. They bought a 12-percentage-point improvement in their odds of walking away from the worst day of their life.

Stack these factors conservatively: ESC full adoption delivers a 15% fleet-wide fatality rate cut; the crossover shift across roughly 18% of the fleet adds another 2–3%; AEB penetration at current levels contributes approximately 5%. Pre-ESC vehicle retirement removing the deadliest model years (2000–2008, which dominated FARS fatalities at rates 2–5 times higher than their modern replacements): compound benefit folded into the ESC figure. Total engineering dividend since 2014: somewhere between 20% and 30%.

Run the math: a 25% reduction from 2014’s 1.08 rate yields 0.81, and a 30% reduction lands at 0.75.

The actual 2025 rate is 1.10, barely better than 2014. At 3.33 trillion VMT, the difference between the expected 0.81 and the observed 1.10 translates to roughly 9,700 additional deaths per year — call it the behavioral tax. Round down for modeling uncertainty and you still get something in the neighborhood of 8,000 people killed annually by a gap between what the cars can do and what the drivers let them.

Where does it go? Three culprits account for the overwhelming majority of lost engineering gains.

Smartphones were the first and largest drain on the engineering dividend. In 2014, 58% of American adults owned one. Today: 92%. NHTSA’s own campaign data says distracted driving kills one person every 2.5 hours and injures 18 every 30 minutes.^[5] The agency is spending $665 million on enforcement grants this year, a meaningful chunk earmarked for “Put the Phone Away or Pay.” For context, the statistical cost of those ~8,000 excess deaths is $100 billion at USDOT’s $12.5 million per life. The $665 million is 0.67% of the damage it attempts to mitigate.

Speed was the second culprit: average travel speeds jumped post-COVID and never fully retreated. Multiple states raised highway limits between 2014 and 2025. Kinetic energy scales with the square of velocity; a 10% speed increase means a 21% increase in crash energy, and while AEB can shave 10–15 mph off an impact, it cannot fix a driver doing 85 in a 65 zone.

Mass disparity was the third and most perverse factor. Average new-vehicle curb weight rose from roughly 4,000 pounds in 2014 to more than 4,400 pounds in 2025, driven partly by the very crossover migration that improved occupant survivability. Heavier vehicles protect their own occupants at the expense of everyone they hit. Our FARS analysis showed a Ram 2500 driver walks away from 79.5% of fatal crashes; a Cavalier driver walks away from 14.3%.^[4] The safety dividend for the person buying the larger vehicle is real, but the math is zero-sum on a population level — somebody absorbs the kinetic energy you exported.

The counterargument

A fair objection: 2014 was anomalously low. It followed the recession-era driving reduction, preceded the rideshare boom, and coincided with moderate gas prices that suppressed recreational driving. Comparing 2025 to a trough inflates the gap. A more honest baseline might be the 2006–2013 average of approximately 1.18 fatalities per 100 million VMT. Against that number, 2025’s 1.10 looks like genuine progress, and the engineering shortfall shrinks from a chasm to a canyon.

It is still a canyon, because even against the higher baseline, the fleet improvements should deliver a rate well below 1.00 by now. The behavioral drag is smaller in this framing but not small.

What the data cannot prove

The engineering dividend calculation requires assumptions about technology penetration rates that FARS cannot directly measure. VMT estimates carry ±2% uncertainty. “Behavioral factors” is a catch-all that absorbs road design deficiencies, EMS response time variation, and changes in driving demographics that have nothing to do with phones or speed. The 8,000-death figure is an estimate with a wide confidence interval, not an actuarial certainty. It is directionally correct, even if it is not precise.

What to do with this

If you are shopping for a car built after 2018 with AEB and side-curtain airbags, the engineering is on your side. Those features represent the most consequential safety improvement since the three-point belt. They work. The limiting factor is you. Specifically your phone, your speed, and your assumption that 4,500 pounds of steel will forgive any lapse in attention — it will forgive yours, but the person in the Corolla next to you is less fortunate.

Regulators should be asking why a decade of extraordinary engineering progress produced a rate change barely outside the margin of error. NHTSA’s $665 million in grants targets symptoms. The FARS data suggests the disease is a driving culture that absorbs safety gains as permission to take more risks, a pattern behavioral economists call risk homeostasis. The cars got better, and we compensated by getting worse. And roughly 8,000 families per year are paying the spread.

Source: NHTSA FARS 2014–2023 (vehicle-specific cross-tabulations); NHTSA 2024 annual and 2025 preliminary estimates. Engineering dividend estimates are modeled from published IIHS effectiveness studies applied to fleet penetration assumptions; see methodology limitations above. See methodology for general caveats.