Fewer road deaths, except on two wheels: thirty years of traffic deaths in New York

Each counts a different thing. The federal record counts a death within thirty days of a crash involving a motor vehicle on a public road; the NYPD’s records count the deaths its crash reports hold, and it doesn’t publish when a later death is added; DOT’s official count applies rules its releases don’t publish. For 2024 the federal record counts 242 deaths, the NYPD’s records 268 and DOT 253. The police count is the highest, and DOT’s releases don’t list which deaths its count leaves out. So we never put two of them on one line, and each comparison stays inside one source.

On the one count we could match, they agree: the NYPD’s records, with each death classed by the dead person’s own vehicle, give exactly DOT’s counts of e-bike, e-scooter and moped riders killed from 1 January to 29 June of 2024 and 2025, and come within one of it in 2023.

These are counts of every death recorded, not samples, so they carry no sampling margin. To judge whether a change could be chance, we treat each count as the outcome of a simple model for rare, unrelated events (a Poisson model) and give an approximate 90% range for the change. These are ranges under that model, not a full measure of uncertainty: they leave out reporting and classification changes. A change is called a rise or a fall only when its whole range lies on one side of zero.

• All traffic deaths, late 1990s to early 2020s: down by between 38.7% and 45.6%; with 2020 and 2021 alone, before the 2022 change, down by between 36.4% and 46.1%.
• Pedestrian deaths, the same years: down by between 44.5% and 53.5%.
• Deaths in cars, vans, trucks and buses, the same years: down by between 57.8% and 66.3%.
• Deaths on two wheels, the same years: up by between 31.2% and 73.0%; with 2020 and 2021 alone, before the 2022 change, up by between 26.8% and 79.9%.
• Deaths on two wheels in the police records, 2020 to 2025 against 2013 to 2019, a year on average: up by between 51.5% and 90.4%; motorcycles alone, up by between 4.3% and 43.4%.
• Deaths on two wheels in the police records, 2023 to 2025: down by between 18.9% and 50.9%; 2025 against the 2013 to 2019 average: up by between 13.2% and 74.2%.
• Deaths recorded on e-bikes, e-scooters and mopeds, 2023 to 2025: down by between 19.5% and 58.8%. On pedal bicycles, 2019 to 2025: down by between 49.9% and 87.5%.
• E-bike, e-scooter and moped riders killed through 29 June, 2026 against 2025, by DOT’s count: anywhere from 36.1% fewer to 98.0% more. “No clear change” means the counts are too small to tell, not that a large change is ruled out.
• Deaths on two wheels in the police records, January to May 2026 against 2025: up by between 3.1% and 209.2% under the simple model, but from 20.2% fewer to 299.5% more once the extra year-to-year swing below is allowed for, so we don’t call it a rise.

The real uncertainty is wider in two ways the model leaves out. Deaths are not quite unrelated events: 31 of the 857 deaths on two wheels from 2013 to 2025 shared a crash with another death, and the yearly counts swing more than the model expects, by up to 2.9 times its variance, estimated from the differences between successive years. Allowing for that swing, every rise and fall the story names still clears zero. Three changes do not, and we don’t call them: e-bike, e-scooter and moped deaths since 2023 (from 67.5% fewer to 2.0% more), motorcycle deaths since 2020 (from 2.1% fewer to 52.7% more) and this year’s first five months on two wheels, above. And the comparisons with 2023 set a later year against a peak picked out of the same counts, which makes a fall look surer than it is.

Before 2022 it filed e-bikes with mopeds, as motorized bicycles, so their riders sat with motorcyclists. From 2022 it counts an e-bike rider as a cyclist and marks whether the bicycle had a motor. Riders of stand-up scooters it has always counted apart, with skates and wheelchairs, as people on personal conveyances. Since 2022 it has also left out crashes involving only e-bikes, since an e-bike is no longer a motor vehicle in its rules (NHTSA’s manual, appendix C).

So the long view keeps motorcycles, mopeds, motor scooters and bicycles together as one line. From 2022 that line leaves out e-bike riders who died with no motor vehicle involved, so if anything it understates the rise; riders of stand-up scooters are outside it in every year. The NYPD’s records name the vehicle and count those crashes, but they begin in 2013.

Not in every year. Until March 2016 a vehicle’s type was chosen from a fixed list with no e-bike on it. After that it is written in, and from 2016 to 2018 no more than 25 vehicles a year were written down as e-bikes or e-scooters, against at least 6,278 bicycles. E-bike and e-scooter labels arrive in numbers only in August 2019; in 2020 the records name 920 e-bikes. So before August 2019 some e-bike riders may be counted as cyclists or moped riders, and we can’t tell how many. That is why we say deaths “recorded” on e-bikes, e-scooters and mopeds.

Two things limit the doubt. Deaths on two wheels in all, and bicycles and e-devices together, don’t depend on telling e-bikes from bicycles, and both rose. And the federal record, which codes each dead person’s vehicle by its own rules and counted e-bikes as motorized bicycles until 2021, also finds few such deaths before 2020: about 2 a year on mopeds, motorized bicycles and motor scooters from 2013 to 2019, against 21 in 2021. It works from the same police reports, though, so an e-bike written up as a bicycle could be missed by both.

The data can’t say. We found no count of how many people ride e-bikes, e-scooters or mopeds, or how far, so these are counts of deaths, not risks per trip. More riders on the road would raise the count even if each ride were no riskier.

At a City Council hearing on e-bikes and e-scooters, as amNY reported on 2 October, DOT’s commissioner said at least two of the 17 e-bike riders killed this year were on illegal devices, with three more unknown, and that six of the e-scooter deaths involved illegal devices, with four unclear. The police records don’t note a device’s speed or power, so we can’t check that or count it for other years.

The NYPD’s open crash records have not been updated since mid-June, and the City’s page for them says the update process is being fixed, so they reach January to May for whole months. The NYPD’s weekly report gives the total to 11 October. DOT’s releases give the count to 29 June, 94 deaths against 91 by the same date of 2025, and, as amNY reported on 2 October, the nine months to September.

The federal record. NHTSA’s Fatality Analysis Reporting System, one file a year from 1995 to 2024, read for the five counties of New York City. A person counts when the injury was fatal within thirty days. The groups come from the person type and, for people in vehicles, the body type of their own vehicle: pedestrians; people in cars, vans, trucks, buses and other motor vehicles, with the few whose vehicle type is unknown; people on two wheels (the record’s motorcyclists, on motorcycles, mopeds, motor scooters and a few three-wheeled and off-road cycles, and its cyclists, e-bikes among them from 2022); and a few others, not drawn (people on personal conveyances, such as stand-up scooters, skates and wheelchairs, and unknown). The file for 2024 is NHTSA’s Annual Report File, which a final file with any late cases replaces about a year later; the earlier years are final. The late 1990s are 1995 to 1999 and the early 2020s 2020 to 2024: five-year averages, since single years of small counts jump about. In every year, the deaths counted here equal the crash file’s own total for the five counties.

The NYPD’s records. The person file, every person recorded killed from 2013 to 9 June 2026, joined to the vehicle file so that each death is classed by the dead person’s own vehicle. The vehicle types are free text, read by a fixed list of patterns: “Standing S” and the code ESST are taken as stand-up e-scooters (“Standing S” for five deaths), mini-bikes and “motorscooter” as mopeds (three mini-bikes), “motorbike” as a motorcycle and a pedicab as a bicycle (one). Every dead person’s vehicle record is read, except for 3 people killed off foot with no vehicle id, who are classed by their crash’s vehicles; none of them was on two wheels. Bicycles, e-bikes, scooters, mopeds and motorcycles come from an extract that holds every vehicle of those types in the City’s vehicle table; the vehicles of the other 860 people killed off foot from 2013 to 2025 were fetched one by one by their ids. Each record belongs to the dead person’s own crash. All the fetched vehicles but 19 are cars, vans, trucks, buses or other motor vehicles; the 19 whose record gives no type (14 of them from 2013 to 2016) are left unclassified. Every death on two wheels from 2013 to 2025 was classed from its own vehicle’s record. Classed instead by the crash’s vehicles, 41 of the 114 e-bike deaths would land in another group.

The rise in the police records. Deaths a year on average from 2020 to 2025 against 2013 to 2019, as the records label the vehicles. To one decimal they rose by 35.6 on e-bikes, e-scooters and mopeds and 6.6 on motorcycles and fell by 7.4 on pedal bicycles: a net 34.8, which rounds to 35, while the parts, each rounded on its own, give 36, 7 and 7.

This year. The NYPD’s CompStat citywide report for the week to 11 October 2026 (traffic fatalities, year to date); DOT’s releases of 2 January and 1 July 2026; amNY’s report of 2 October 2026 on DOT’s nine-month figures.

Chance. See “How sure are these changes?” above. Under the same model, this year’s total against last year’s runs from 21% fewer to 16% more: about the same. A year set against a several-year average is compared as a rate a year.

Code and checks. One script, analysis.py (linked below), produces every figure and chart from these files and checks each figure it quotes against the saved copy of its report. In data.csv, the federal record’s two-wheel line is its motorcyclists and cyclists added together. The dataset cards behind it (fars_nyc, nypd_collisions, nypd_crime_history) list the traps.