Safe Streets Map
MethodologySources, methods and limits

How the data works

Everything on Safe Streets Map comes from public records. This page explains where they come from, how we turn them into hotspots, and what they can't tell you. Data last rebuilt 2026-10-22.

In numbers

  • 64,774 pedestrian crashes off freeways, 2021–2026: 17,478 serious or fatal, 4,975 fatal.
  • 58,317 bicycle crashes, 2021–2026: 7,848 serious or fatal, 973 fatal.
  • City pages for 334 cities (walking) and 332 cities (cycling).

Where the crash records come from

We don't collect, create or correct crash records. They are government records, written and published by the agencies below; we download them and show our own reading of them. A mistake or a gap in a police report stays in our maps and pages. These agencies don't make, check or endorse this site, and nothing here is an official record. For a report about a particular crash, ask the agency that took it (see how to get a police report).

  • Crashes: California Crash Reporting System (CCRS), published by the California Highway Patrol on the state's open data portal, data.ca.gov, which lists it as updated daily. It is built from the reports officers of the CHP and local police departments fill in at crashes. Reports keep arriving for months, so the latest year is incomplete and earlier years can still change. We use the crash, party and injured-person tables: where and when each crash happened, how badly people were hurt, what kind of collision it was, what the person on foot was doing, and whether a driver was turning or backing up.
  • Deaths: Fatality Analysis Reporting System (FARS), from the National Highway Traffic Safety Administration (NHTSA). Analysts working for a state agency in each state put each case together under an agreement with NHTSA. A crash is included when it involves a motor vehicle on a road open to the public and someone dies within 30 days. NHTSA first releases a year as an Annual Report File, then replaces it about a year later with a Final File that adds cases and updates others, so the newest year's deaths can still change.
  • Streets, paths and bike lanes: OpenStreetMap contributors, shown through Mapbox.

How hotspots are found

  1. Every crash with map coordinates is placed on a grid of 150 m squares.
  2. Starting from the busiest square, neighbouring squares within 250 m are merged into one hotspot, so each hotspot centres on the worst spot.
  3. A hotspot needs at least 3 crashes. Fewer than that is too little to call a pattern.
  4. Hotspots are ranked by harm, not just count: each crash counts 1, a crash where someone was seriously hurt counts 3, and a crash where someone died counts 6.
  5. Each hotspot is named after the streets its crashes were reported on, and described from what the reports say happened, for example how many people were hit by a turning driver, or how many crashes happened after dark.

Pedestrian crashes on freeways are left out of the walking map and city pages, since they're mostly people stranded or walking on the freeway rather than crossing streets. Cities need at least 15 reported crashes to get a page.

Where the numbers on The Harm Cars Do come from

The The Harm Cars Do page counts deaths from FARS (2021–2024), crashes that injured someone from CCRS, and land from OpenStreetMap and the US Census. Its map card adds the Census American Community Survey (2020–2024 five-year estimates) for population, commute time, how people get to work and households without a car. Cost and mode comparisons are national figures from the IRS, the Bureau of Labor Statistics, NHTSA, the NTSB, the CDC, the Federal Highway Administration and the Federal Transit Administration, each shown with its year. Deaths and crashes are divided by residents, not by miles driven, and a city's deaths include people passing through. Road area is an estimate from road length and lane width, and parking is only what OpenStreetMap volunteers have mapped, so both are minimums.

How the findings are found

The findings come from a script that tests thousands of slices of the crash records: how often a kind of crash kills, how a place's mix of crashes differs from everywhere else's, and the weeks around the November clock change. It runs on 2021–2023 (2021–2022 for deaths outside California), keeps only patterns that pass a Benjamini–Hochberg false-discovery test and a minimum size, and keeps those only if they hold in the later years. It must first rediscover patterns already known. A person then drops patterns that come from how agencies record crashes, such as speed limits or hit-and-runs by city. Deaths FARS added to the map are left out of these tests, because their reports carry fewer details.

What the numbers leave out

  • Crashes without coordinates. Many city police departments send reports with street names but no map location (in Los Angeles, San Francisco, San Diego, Sacramento and Oakland, under a tenth of pedestrian and bike reports have one). We place those at the intersection the report names, found in OpenStreetMap within the city, and flag them in the popup; where the streets are offset by a block or more the point is still the intersection. Reports whose streets can't be found (freeways named by route number, alleys, parking lots) aren't shown, and each city page gives the share of its reports that are on the map. Neighborhood pages count a crash toward the city-drawn outline that holds its point, so a placed crash near a border can fall on either side.
  • Deaths are nearly complete; injuries are not. Fatal crashes come from NHTSA's Fatality Analysis Reporting System (FARS), the federal count of deaths in crashes, and are added to the state data. As a check, San Francisco's own injury-crash records for 2021–2025 list 78 pedestrian deaths and 4 cyclist deaths; this map has 77 and 3. The same records list 2,920 pedestrian and 1,716 cyclist crashes with an injury, and this map has 131 and 56, because most San Francisco reports in the state data have no map location. Expect the same gap in other cities that send reports without coordinates.
  • Other states show deaths only. Outside California, the walking, cycling and motorcycling maps show FARS fatal crashes (2021–2024) with a dark ring, and no injury crashes, so other states look emptier than California. They are left out of the crash glow and the hotspots, and the maps' other layers (weather, closures, riding roads, rider stops) cover California only.
  • Unreported crashes. Many crashes, especially minor bike crashes, never get a police report.
  • How many people use a street. A busy street shows more crashes partly because more people walk or ride there. The counts show where crashes happen, not the risk per trip.
  • Recent changes. A street redesigned after 2026 may be safer than its history suggests. Newer crashes are added each time the data is rebuilt.

Data sources and licences

Every source the site uses, what it's used for, and the terms it comes under.

Files we derive from OpenStreetMap (riding roads, twisty roads, climbs, ride lines, places) are available under the same ODbL. Map tiles, static map images and route directions come from Mapbox, a paid service under its own terms, not open data.

Using and citing the data

You're welcome to quote the numbers and link to any page. Please cite it as: “Safe Streets Map analysis of California Crash Reporting System data, 2021–2026” and link to the page you used. For corrections or data questions, email contact@safestreetsmap.com.