Crash Atlas · Shared methodology
Crash Atlas methodology
Every Crash Atlas edition follows the same data process. This page is the long-form reference. Individual editions include an edition-specific methodology block with the source dataset, date range, and any local quirks.
What we publish
Each Crash Atlas edition is a public-data study of truck-involved fatal and severe-injury collisions in one US metro or county over a defined date range. The published artifacts include:
- Headline aggregate numbers (deaths, severe injuries, pedestrian and bicyclist victims)
- A geocoded crash map with severity and time filters
- Year-over-year trend
- Top corridors (state and federal highways)
- Hour-of-day distribution
- By-city or by-ZIP breakdown
- Truck-type and victim-demographic detail
- Raw CSV + JSON downloads under CC BY 4.0
Data sources
Crash records come from the authoritative public dataset for each state. In California that's SWITRS (Statewide Integrated Traffic Records System), accessed via the UC Berkeley TIMS portal which applies consistent geocoding. Other states use their own equivalents (Texas C.R.I.S., Florida FCSR, Illinois IDOT, Georgia GEARS, New York DMV). National coverage is supplemented with the federal Fatality Analysis Reporting System (FARS) from NHTSA where state datasets lag or have gaps.
Truck-involved definition
"Truck-involved" means a collision in which at least one party was a large truck. The exact filter depends on the source dataset:
- SWITRS (California): we use the SWITRS TRUCK_ACCIDENT flag at the collision level, which the California Highway Patrol applies during record processing. This predominantly captures big-rig tractors (with or without trailer) but can include some pickup-with-trailer combinations.
- FARS (federal): body_type codes 60-79 (Single-Unit Truck) and 80-89 (Truck Tractor / Combination).
- Other state datasets: documented per-edition.
Where the source dataset supports it, we publish a party-level vehicle-type breakdown so readers can see the actual mix (big rig with trailer, big rig without trailer, pickup with trailer, bus, etc.).
Severity filter
Headline numbers are restricted to KSI — Killed or Severely Injured. Lower severities (visible injury, complaint of pain, property-damage-only) are known to be inconsistently reported across police agencies and are noise for the corridor-pattern analyses that make these studies useful. Total lower-severity collision counts are reported in each edition for context.
Geocoding
We use the geocoded coordinates supplied by the source dataset (e.g. TIMS' POINT_X and POINT_Y for SWITRS). Each edition reports what percentage of KSI records had valid coordinates. We exclude points that fall outside the metro's bounding box — these are geocoding errors that would otherwise pollute the corridor and heatmap analyses.
Date range and preliminary data
Each state dataset has its own publication lag. SWITRS finalises records about 12-18 months after the calendar year ends, with preliminary data available sooner. Each Crash Atlas edition reports the date range used and explicitly marks any year that's still preliminary (preliminary years are not included in the headline "Vision Zero baseline" comparison if the host city has one).
What we don't try to measure
- Per-driver fault. SWITRS and most state datasets record an "at fault" flag, but the standard is "primary collision factor" determined by the responding officer — not a court finding. Our headline numbers do not attribute fault.
- Carrier-level safety scores. Carrier safety is a separate dataset (FMCSA SAFER); when relevant we link out rather than try to merge.
- Settlement or verdict outcomes. SWITRS records crashes, not lawsuits. Outcomes of any civil litigation that may follow a crash are out of scope.
- Causation. A correlation between a corridor and a high crash count does not establish that the corridor design caused the crashes. We report the pattern; interpretation is up to readers and reporters.
Reproducibility
The cleaning and aggregation script for each edition is published alongside the raw download. Anyone with access to the underlying state dataset can re-derive the same numbers from the same source. We document the exact filter values, column names, and severity mappings so a reproduction takes minutes, not hours.
Updates and corrections
Each edition has an "Update log" section. Material corrections (a number changes by more than 1%, a methodology decision is revised) are noted there with the date. Editions are refreshed when the source dataset finalises preliminary years or publishes a new year.
Methodology last updated: 2026-05-25. The Crash Atlas series is compiled by the AccidentLawyerReview Editorial Team.