The 2014 Ghost: Why the Safest Year in Driving History Was 11 Years Ago

Eleven years. That is how long the safest driving year in American history has been dead, undisturbed, collecting dust on a shelf at NHTSA while the entire automotive industry pretended the next sensor suite would finally surpass it.

In 2014, the United States recorded 32,744 traffic fatalities at a rate of 1.08 per 100 million vehicle miles traveled, a number so low it has not been equaled since.^[1] In 2025, after a decade that gave us automatic emergency braking on 95% of new vehicles, mandatory backup cameras, lane keeping assist, blind spot monitoring, and near-universal electronic stability control fleet penetration, NHTSA's preliminary estimate landed at 1.10 per 100M VMT.^[2]

We are further from 2014's record than we were in 2019.

Do the arithmetic yourself: Americans drove approximately 3.33 trillion miles in 2025. At 2014's rate, that mileage would have produced roughly 35,964 deaths instead of the estimated 36,640.^[1][2] The difference is 676 bodies per year, 56 per month, 13 per week, almost two per day, accumulating silently in the margin between what the country once accomplished and what eleven years of technological progress apparently cannot replicate.

The 2014 miracle had one primary engine: electronic stability control. FMVSS 126 required ESC on all new passenger vehicles starting with model year 2012, and by 2014 the fleet penetration curve had reached the steep part of its climb.^[3] IIHS estimated ESC reduced single-vehicle crashes by 33% in cars and 56% in SUVs.^[4] NHTSA's Charles Kahane calculated that ESC saved 1,575 lives in 2014 alone, a number that would grow to 1,949 by 2015 as more pre-2012 vehicles rotated out of the fleet.^[5] One technology, invisible to the driver, requiring no app and no conscious input from anyone behind the wheel.

Everything added since should have pushed the rate lower. Backup cameras became mandatory in May 2018, preventing an estimated 58 backover deaths per year.^[6] Twenty automakers committed voluntarily to AEB in 2016, and by 2023 roughly 95% of new vehicles shipped with it; IIHS found AEB cuts front-to-rear crashes by 50%.^[4] Lane departure warning reduces relevant crash types by 11%. Blind spot detection cuts lane-change crashes 14%.^[4] Stack those percentages and you would expect a rate comfortably below 1.00 by now, perhaps approaching 0.90, a number that would save more than 6,000 additional lives per year at current VMT levels.

Instead: 1.10, still losing to 2014 by two hundredths of a point, which raises the question of where exactly the technology dividend went. Pedestrians absorbed most of it. In 2014, 4,884 pedestrians were killed by motor vehicles. By 2022, that number hit 7,522, a 54% increase that represents 2,638 additional deaths per year in a single crash category.^[7] If pedestrian fatalities had held flat at 2014 levels while every other category improved as the technology promised, the overall rate would have broken 1.08 years ago. Pedestrian deaths went from 15% of all traffic fatalities in 2014 to 18% in 2022, and the geometry of the problem is simple: vehicles got taller, hoods got higher, and humans walking near roads became proportionally more fragile relative to the machines that struck them.^[8]

Apple released the iPhone 6 in September 2014, and within two years large-screen smartphones were universal, Instagram Stories launched, and ride-share apps placed constant navigation demands on drivers who had never previously been professional route-finders. The fatality rate spiked to 1.18 in 2016 and has never returned to 2014 levels. NHTSA attributed 3,308 deaths to distraction in 2022, but that figure captures only cases where an officer documented direct evidence of phone use at the scene, which is roughly as diagnostic as detecting intoxication by asking the driver how many drinks they had.^[9]

Vehicle mass crept upward too. Average new-vehicle curb weight rose from roughly 3,900 pounds in 2014 to approximately 4,300 pounds in 2024, a gain of 400 pounds absorbed primarily by the shift from sedans to SUVs and pickups. SUV market share climbed from 36% to 57% of new sales over the same period. IIHS has demonstrated that every 1,000-pound increase in the striking vehicle raises the other vehicle's occupant fatality risk by 47%.^[8] The fleet got safer for the people inside it and more lethal for everyone outside.

Then 2020 happened and the pandemic emptied the roads. VMT dropped. Fatality rates spiked to 1.34 as the remaining drivers treated open highways like private racetracks, and the behavioral habits formed during lockdown (faster speeds, less belt use, more impaired driving) did not fully reverse when traffic returned.^[1] The rate peaked at 1.37 in 2021, levels not seen since 2005, and has been clawing back toward 2014 ever since, reaching 1.19 in both 2023 and 2024 before the preliminary 2025 estimate of 1.10.^[2]

So 2025 is the second-safest year in history, and you can celebrate if you want, but understand what the eleven-year gap means: the cumulative technology investment since 2014 produced a net safety improvement, across the entire national fleet, of approximately zero. ESC saved thousands, AEB saved thousands, and backup cameras, lane keeping, blind spot monitoring, and improved crash structures all saved real lives too. And then bigger vehicles, faster phones, heavier curb weights, taller hoods, and residual pandemic recklessness consumed every single one of those saved lives and demanded more.

What to do with this

If you are buying a car: check IIHS pedestrian-crash ratings, not just occupant protection. A vehicle with top marks for the person inside it and no pedestrian AEB is a vehicle optimized for the wrong half of the problem. If you own a phone: enable Do Not Disturb While Driving (iPhone: Settings, Focus, Driving; Android: Digital Wellbeing, Driving Mode). If you are a policymaker: the next 10,000 lives will not come from adding more sensors to vehicles that already have twelve of them. They will come from redesigning front-end geometry to reduce pedestrian lethality, enforcing speed limits with infrastructure rather than honor systems, and acknowledging that the 2014 rate was not a floor to build on but a ceiling we have failed, for eleven consecutive years, to break through.

Limitations

The 2025 fatality data is preliminary. NHTSA's final FARS count historically deviates from early estimates by 1 to 3 percent in either direction, which could shift the 2025 rate above or below 1.10. VMT estimation methodology has evolved since 2014, and small denominator adjustments across trillions of miles produce visible rate changes even when the underlying reality is stable. This analysis cannot cleanly isolate how much each behavioral factor (distraction, speed, vehicle size) individually contributed to consuming the technology gains because FARS records crash outcomes, not the chain of pre-crash decisions that produced them. ESC lives-saved estimates rely on counterfactual modeling comparing actual crash frequencies against projected frequencies without ESC, not direct observation of crashes prevented. FARS captures fatalities only; we cannot assess whether the technology investment reduced injury severity even as it failed to reduce fatality rates.

Strongest counterargument

A 2025 VMT mile is not a 2014 VMT mile. The composition of driving has fundamentally shifted: more urban delivery trips from DoorDash, Amazon, and Instacart drivers navigating unfamiliar residential streets with tight deadlines; more ride-share VMT from Uber and Lyft drivers operating in dense mixed-traffic environments; more e-bike and e-scooter traffic creating complex interaction geometries that did not exist in 2014. If the average 2025 mile is inherently more dangerous than the average 2014 mile due to these compositional changes, then achieving a rate of 1.10 under harder conditions might represent a larger real safety improvement than the raw number suggests. The technology may have worked, but we cannot see it because the denominator changed underneath us.