This page explains, step by step, how every figure on NY Crash Records is produced. If you want to check a number, you can repeat the same calculation on the public dataset.
The source
All statistics come from the Motor Vehicle Collisions – Crashes dataset published by the City of New York on NYC Open Data. Each row is one crash reported by the NYPD.
A police report (form MV-104AN) is required when someone is injured or killed, or when damage reaches $1,000. Minor crashes that nobody reported to the police are not in the data.
The dataset in its current form starts in July 2012. We use full years from 2013 onward.
How often the figures are updated
The City updates the dataset regularly. We rebuild every page once a month. The date the data runs through is shown at the top of every page and in the source line under each set of figures.
Recent months are preliminary. Reports can arrive late or be corrected, so totals for the latest months may rise slightly after we publish them.
The time periods we use
- Recent period: the main totals and rankings on each page cover the last three full calendar years. Using three years smooths out random ups and downs at small locations.
- Yearly charts: show every full year since 2013 (since five years ago for streets) plus the current year so far, drawn with a hatched pattern because it is incomplete.
- Year to date: counts crashes from January 1 through the last date in the data.
What we count
- Crashes: the number of police reports.
- Injured and killed: the number of people listed as injured or killed in those reports, split into pedestrians, cyclists and motorists (drivers and passengers). One crash can injure several people, so injuries can exceed crashes.
- Injuries per 100 crashes: people injured divided by crashes, times 100. It shows how often crashes at a place hurt someone, not how many crashes there are.
Boroughs and ZIP codes
We use the borough and ZIP code written in each report. A large share of reports — roughly three in ten in recent years — have no borough or ZIP code, mostly crashes on highways and expressways. Those crashes count in the citywide totals but cannot be assigned to a borough or ZIP code.
For that reason, local totals are a floor, not an exact count. Each page shows the share of crashes with no borough for the same period.
We publish a ZIP code page when it has at least 100 crashes in the recent period.
Streets and intersections
Officers type street names by hand, so the same street appears in many spellings ("W 42 ST", "WEST 42 STREET", "West 42nd St"). We standardize common abbreviations, directions and ordinal numbers so those variants count together. Some unusual spellings may still be missed.
A crash belongs to a street when the report names it as the street where the crash happened, and to an intersection when the report also names the cross street. Crashes logged only with coordinates or with a street address (for example "280 St Nicholas Ave") are counted in borough and ZIP code totals, but not on a street or intersection page.
Thresholds for a page: streets need at least 150 crashes and intersections at least 30 crashes in the recent period. Rankings compare places that meet the same threshold.
Causes and vehicle types
Officers can list up to five contributing factors, one per vehicle. We use the factor recorded for the first vehicle, which is usually the main one. Reports marked "Unspecified" are counted separately and excluded from percentages of known causes.
A contributing factor is the officer's judgment at the scene. It is not a finding of legal fault and can differ from the outcome of an insurance claim or a court case.
Vehicle types are written in hundreds of different ways. We group them into broad categories (passenger car, SUV, taxi, bus, box truck, e-bike, and so on), again using the first vehicle listed.
What the data cannot tell you
- It does not measure risk per driver or per mile. Busy streets record more crashes partly because they carry more traffic.
- It does not include crashes that were not reported to the police.
- It does not identify the people involved. We never publish names, plate numbers or any personal details.
- It cannot tell you who was at fault in a specific crash.
Spotting an error
If a figure looks wrong or a street name has been merged incorrectly, please let us know. We review every report and correct the calculation for all pages at the next monthly update.
Last updated: November 4, 2026