Per subway ride, major felonies fell for thirty years. Felony assault broke from the fall.

• Felony assault up 11% this year. NY1’s figure (14 October) is a count through September, with major crime overall up less than 1%; the detailed comparisons stop at May. In those months the records hold 272 felony assaults against 267 in the same months of 2025, and per entry 1% lower, between 14% lower and 14% higher under the model used throughout: no clear direction.
• Subway crime about a fifth higher per ride than before the pandemic. Vital City’s reading (August; 1,071 major felonies in the first half of 2026 across nearly 650 million rides) divides by the MTA’s ride counts, which leave fare evaders out, over January to June. Our January to May comparison on the same paid-ride basis gives 22% more than 2019 (between 13% and 31%): close, over different months. Count every entry, paid or not, and it is 13% (5% to 22%) in those months, and over the year to June 2026 the six were about level with 2019. Neither cancels the other: the shorter window shows a rise, the fiscal year does not.
• Felony assaults and grand larcenies drive transit crime. The NYPD’s third-quarter release puts them at 30.7% and 49.3% of this year’s transit crime. In the detailed records felony assault was 27% of the six in 2025 and grand larceny the most common, so the shares agree in kind; the NYPD’s count is the transit system’s year to date and ours the subway’s calendar year, and a large share says nothing about what moved the total.
• Subway violence up by a sixth since the pandemic and nearly doubled in a decade. Vital City’s “violent crimes” are murder, robbery, rape and felony and misdemeanor assault together, over calendar years to 2024, so its measure holds the misdemeanour assaults ours leaves out. Our felony assault alone was 2.1 times its 2019 rate in January to May 2026 and robbery showed no clear direction: both point the same way, but the measures and the dates differ, so this is context, not a check.
• The second safest year in fifteen. The NYPD’s statement is its description of recorded counts, with the pandemic years left out; we don’t re-rank the years here. Per ride, counting fare evaders, the year to June 2026 was about level with 2019, and this year’s count comes with fewer riders than 2019 had.

The MTA counts entries made with a fare card or a tap, each time someone swipes or taps in; people who enter without paying aren’t counted. The MTA estimates their share each quarter by having checkers watch the gates: about 4% of entries in January to May 2019 and 10% in the same months of 2025. Left alone, that rise would make 2025’s months look six or seven per cent worse per ride than 2019’s. So for the comparisons with 2019 we count every entry, an estimate: each month’s paid rides divided by the share that paid in its quarter, taking each quarter’s estimate to hold through its months. The long view and the second chart use paid rides, since the MTA’s estimates begin in 2018; the third counts every entry.

The MTA’s daily counts begin in March 2020, so 2019’s months are rebuilt: its official total for the year, spread over the months by the pattern of rides in the MTA’s comparisons of each 2024 day with its 2019 equivalent, leaving out 29 February 2024. January and February 2020 together are 2020’s official total less the daily counts from March. In 2021 to 2024 the daily counts come within 0.2% of the official totals. Rebuilt from those comparisons as they stand, with the leap day kept, every comparison with 2019 comes out the same; both rest on the same comparisons, so this checks the method, not the months themselves.

The weekly report and the Mayor’s Management Report count the crimes the NYPD’s Transit Bureau handles anywhere in the transit system. The NYPD’s detailed records, which name each crime and the month it happened, count those whose place is the subway. For the year to June 2026 the Mayor’s Management Report counts 2,176 major felonies and the detailed records 2,093, about 4% fewer; the detailed records’ June is incomplete and their late reports are left out, so this is a check that the two are close, not a match. The detailed records also hold no murder with the subway as its place, so their six are in practice five; the transit system’s few murders are in the Mayor’s Management Report’s count. Each comparison stays inside one source.

Reports and records. A crime nobody reports is not counted, and how many crimes are recorded depends partly on how many officers are in the subway and what they look for; the NYPD calls each record a complaint.

Felony assault is also a legal category the State has widened over the years. Strangulation has had a felony code of its own since November 2010, and the law has added more kinds of transit and public worker, most recently in September 2022 and July 2024; assault on a retail worker became a felony in October 2024 and has no code of its own in the complaint files. Newly covered transit workers may sit in the code for other public service employees or in the general one, which is why leaving out both worker codes only narrows the question.

We follow six serious crimes that victims mostly report, and leave out the offences the police mostly find themselves, such as weapons, drugs and fare evasion, whose counts can follow enforcement as much as offending; the theft-of-services jump in the story is the example. The files renamed that offence’s category over the years while its code stayed the same, so we count by the code.

The Mayor’s Management Report prints the transit system’s major felonies in this form from the year to June 1998; that first year’s figure comes from the fiscal 2001 edition, which marks it as based on preliminary data. Earlier counts survive in news reports and secondary studies with other definitions, so we don’t join them to this line. The rides for each fiscal year are added up month by month: up to 2019, each calendar year’s official total spread by 2019’s pattern of rides, an assumption; January and February 2020, 2020’s official total less the daily counts from March; from March 2020, the daily counts alone. The six offences are counted under each year’s law, and the changes we found widen them.

These are counts of every crime recorded, not samples. 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) over its number of rides, and give an approximate 90% range, taking the rides as known. The model leaves out changes in reporting and policing, the margins of the fare-evasion estimates and extra year-to-year swings, so the true uncertainty is wider. A change is called a rise or a fall only when its whole range lies on one side of zero.

• Major felonies per ride, the year to June 1998 against the year to June 2026: down by between 65.3% and 68.1%.
• Major felonies per ride, January to May 2026 against the same months of 2025 (876 against 907 complaints): between 13% lower and 1% higher. With 2026’s unpaid share a point above the MTA’s estimate and 2025’s a point below, it would read as a fall.
• Felony assault per ride, January to May 2026 against 2019: up by between 79% and 149%. Leaving out the codes for assaults on police and on other public service employees, between 1.6 and 2.4 times its 2019 rate.
• Felony assault per ride, January to August 2022 against the same months of 2019, before any of the law’s widenings: between 2.2 and 2.9 times the 2019 rate.
• Grand larceny, 593 complaints in January to May 2019 and 421 in the same months of 2026: down per entry by between 4% and 22%; on paid rides alone, between 16% lower and 3% higher. Moving the MTA’s evasion estimates a point against the fall, it is still down, by between 2% and 20%. Allowing for the extra month-to-month swing of thefts, though, the January-to-May range runs from 27% lower to 3% higher, so we call it a gap that could still be chance. Over the whole of 2025 against 2019 (1,131 complaints against 1,572), counting every entry, it is down by between 4% and 16%, and by between 2% and 18% with that swing allowed for: the fall that holds.
• Robbery: between 4% lower and 36% higher; on paid rides alone, up by between 3% and 46%.
• The six major felonies together, January to May (876 complaints in 2026 against 942 in 2019): up per entry by between 5% and 22%. Over the whole year to June 2026, in the Mayor’s Management Report’s count (2,176, against 2,590 in the year to June 2019) and counting every entry, they were about level with the year to June 2019.

Each comparison we name with 2019 or with last year is also run again three ways (the long view against 1998, a fall of two thirds, is not): with the MTA’s fare-evasion estimates moved a point against it, with 2019’s months rebuilt the other way, and with late reports added as they ran in 2025. The first is a fixed scenario, not the estimates’ own margins: the MTA publishes margins of 0.8 to 1.0 points for the quarters compared, at a confidence level it doesn’t state, and none for 2019. Every comparison keeps its direction under each of them, except the six against last year, noted above.

Detailed records. The NYPD’s complaint records, historic and year to date, counted on the City’s server for complaints whose premises is the subway, by the month the crime happened, from January 2014 to June 2026. June 2026 is incomplete, so this year’s comparisons use January to May, and the 109 complaints filed in 2026 about earlier months are left out. Some crimes are reported late: about 5% of December 2025’s felonies were filed in the new year. The year-to-date file holds reports up to 30 June 2026, so May has had a month to be reported, as November 2025 had by the end of that year. If each month’s late reports run as they did for the same position in 2025, January to May 2026 would be about 0.7% short of its count six months on, which would change none of the comparisons named here; this is a scenario, not a bound.

Rides. Paid subway entries: New York City Transit’s annual counts to 2013 as NYU’s Rudin Center tabulated them, the MTA’s yearly figures from 2014, and the MTA’s daily counts from March 2020, built month by month as described above. By calendar year, in the detailed records, the six major felonies ran at 1.47 per million paid rides in 2019, 2.72 in 2020 and 1.66 in 2025, close to the fiscal-year rates of the long view; felony assault was 0.61 in 2021 and 0.46 in 2025.

Chance. See “How sure are these changes?” above.

Code and checks. One script, analysis.py (linked below), produces every figure and chart from these files, checks each figure it quotes, and a passage of each law it cites, against the saved copies of their sources, and runs the comparisons with 2019 and with last year under the what-ifs above; its results file lists which applies to which. The dataset cards behind it (nypd_subway_complaints, nypd_crime_history, mta_ridership) list the traps.