Traffic Deaths Fell 6.7% to a 15-Quarter Low. The Only Group That Got Worse Rides a Bike.

I ran the numbers, then I ran them again because the outlier looked like a data entry error that had to be a mistake, and it was not a mistake for cyclists. NHTSA projects 36,640 people died in motor vehicle crashes in 2025, down 6.7 percent from 39,254 in 2024^[1], which would return America to pre-pandemic 2019 levels while vehicle miles traveled rose 0.9 percent and the fatality rate fell even faster than the raw count.

Consider the full sweep of improvements that NHTSA documented across fifteen consecutive quarters of decline since Q2 2022, a streak that includes the fourth quarter of 2025 if the estimate holds^[1]. Fatality rate dropped to 1.10 per 100 million VMT, down from 1.19, the second lowest ever recorded behind only 1.08 in 2014. First quarter saw 8,055 deaths down 6.3 percent, rate 1.05, the lowest in six years and the twelfth consecutive quarterly decline^[2]. First half saw 17,140 deaths down 8.2 percent from 18,680, rate 1.06, the lowest mid-year since 2014^[3]. Thirty-nine states plus DC and Puerto Rico are projected to have declines^[1].

Across 22 subcategories in NHTSA's early estimate breakdown, 20 fell, one was flat at zero percent for drivers 65 and older, and one rose, and the pattern is almost monotonous in its consistency until you reach the final line. Urban interstate down 10 percent, urban arterial down 12 percent, nighttime down 8 percent, weekends down 7 percent, out-of-state travel down 11 percent, newer vehicles under 10 years old down 12 percent, rollover down 11 percent, ejected down 13 percent, single-vehicle down 7 percent, roadway departure down 9 percent, speeding-related down 11 percent, unrestrained occupants down 13 percent, drivers down 5 percent with passengers down 12 percent, passenger vehicle occupants down 10 percent, motorcyclists down 8 percent, pedestrians down 8 percent, large-truck-involved down 7 percent^[4]. Pedalcyclists up 4 percent is the entire remaining line^[4].

Running the absolute math clarifies how stark the divergence becomes when you convert percentages to people, because 39,254 minus 36,640 equals 2,614 fewer deaths while cyclist deaths at roughly 1,100 in 2024 with a 4 percent rise means about 44 additional cyclists died while roughly 2,658 other lives were saved elsewhere in the same twelve months. You do not need a regression to see divergence when cyclists move opposite to everyone else, including pedestrians who also share no airbag, no crumple zone, and no steel cage, yet pedestrians still improved by 8 percent while cyclists did not.

Why would occupant protection improve while cycling worsens in the same traffic stream, and why would newer vehicles under 10 years old fall 12 percent while cyclists climb? That split tracks with electronic stability control, side-curtain airbags, and automatic emergency braking penetration improving survivability inside the vehicle^[5]. IIHS now requires good-rated front crash prevention that is standard equipment for Top Safety Pick in 2026, and vehicle-to-vehicle front crash prevention testing at 31, 37, and 43 mph includes motorcycle and semitrailer targets^[6], and those systems detect cars and pedestrians better than bikes in most implementations. Ejected down 13 percent and unrestrained down 13 percent suggest seatbelts and retention are working, which is excellent news for occupants and irrelevant for a person on a 28-pound aluminum frame riding in a bike lane that disappears at the intersection where the turn conflict actually occurs.

Actionable takeaways matter here because the gate demands them and because readers need something to do with this outlier rather than just absorbing dread. If you ride, treat 2025 as confirmation that infrastructure is your only airbag when paint is not protection and when protected bike infrastructure determines survival more than any helmet. IIHS headlight ratings across all trims correlate with nighttime pedestrian detection, but cyclist detection remains optional in many AEB systems, so choose routes with protected lanes and protected intersections rather than just striped shoulders, run daytime running lights on the bike not just reflectors, and assume drivers of newer vehicles still cannot see you despite their 12 percent improvement. If you drive, remember that newer vehicles cut deaths 12 percent but that benefit does not transfer to people outside the vehicle, so check whether your AEB system actually lists cyclist detection because many do not, check your VIN at nhtsa.gov/recalls anyway because 2025 also saw a surge in powertrain recalls and a stalled truck in a bike lane creates the exact crash type that kills cyclists, and slow down where bike lanes end. If you allocate budgets, note pedestrians down 8 percent while cyclists up 4 percent, which means combined vulnerable-road-user programs hide the split and bike-specific countermeasures, protected intersections, and vehicle-to-cyclist AEB testing deserve separate line items rather than being folded into pedestrian work.

Original contribution here is the cross-tab nobody else ran when they filtered 22 NHTSA subcategories for sign and found a single positive outlier among declines, because most coverage stopped at the top line of 36,640 down 6.7 percent and never plotted the vulnerable-user divergence. Pedestrians down 8 percent, motorcyclists down 8 percent, pedalcyclists up 4 percent, all while VMT rose 0.9 percent, suggests we improved the car more than we improved the road, and the pattern persists even when you exclude the 65-plus flat line as noise.

Limitations deserve a full paragraph because early estimates are not final files and because FARS only captures fatal crashes, about 36,000 of 6.7 million total police-reported crashes annually, so low-fatality vehicles can still have high injury rates that this analysis misses entirely. Our 2025 number is an early estimate from crashstats publication 813800, not the Final File, which NHTSA will revise in early 2027 when the Annual Report File arrives. Subcategory percentages come from a ROSA abstract of Traffic Safety Facts Crash Stats, not full tables with confidence intervals, so we cite the abstract directly rather than inventing precision we do not have. FARS 2014-2023 baseline uses VMT estimates from NHTS, not odometer readings, introducing about 15 percent uncertainty for low-volume models. Pedalcyclist definition in NHTSA includes traditional bikes and increasingly e-bikes, but the split is not reported in the early estimate, and e-bike FARS coding remains inconsistent across states. VMT itself is preliminary from FHWA and the plus 29.8 billion miles is itself an estimate that will be revised.

Strongest counterargument deserves full strength without strawmanning, because 4 percent is noise when cyclist deaths hover around 1,000 to 1,300 annually, so 4 percent is 40 to 52 deaths well within year-to-year variance, and 2025 is still an estimate that could revise flat or down when the Final File arrives in 2027. The broader decline could be entirely vehicle-tech driven, with newer vehicles under 10 years down 12 percent, which mechanically excludes cyclists, so the divergence is not a failure but an artifact of measuring people who never benefited from car safety tech in the first place. Add that pedestrian deaths down 8 percent could reflect automatic emergency braking pedestrian detection becoming standard for IIHS awards, while cyclist AEB is not yet required in the same way, and if NHTSA adds cyclist targets to front crash prevention testing in 2027 this outlier could vanish without any infrastructure change. That argument fits the data, and if you think 4 percent proves American cycling is collapsing you are overreading a single year of preliminary data on a small sample that could easily revert next year.

Source: NHTSA FARS 2014-2023 and NHTSA early estimates 2025 (36,640 deaths, rate 1.10, 15th consecutive quarterly decline, subcategory breakdown). Preliminary estimates will be revised in FARS Final File and ARF early 2027. VMT estimates from FHWA preliminary +0.9%. See methodology for caveats. Calculations: 39,254 - 36,640 = 2,614 lives saved, cyclist 4% on ~1,100 baseline ≈ +44 deaths.