Lincoln

Methodology & data quality

Dashboard

Data integrity

What the dashboard measures—and what it does not.

California Collision Intelligence analyzes public California Highway Patrol CCRS records while preserving the distinctions between collisions, involved parties, and people.

33

annual source files

143

source fields modeled

2016–2026

available crash years

77.9%

2026 map coverage

Source coverage

CCRS publishes three tables per year: collisions, involved parties or vehicles, and people. The dashboard inventories every published field and builds period-specific analytical snapshots from the official files and API metadata.

Every dashboard page can select any year from 2016 through 2026, or combine the full range with “All years.” The state API’s 2019 crash index currently ends in April, so the dashboard rebuilds that year’s statewide totals, map, and analytical dimensions from the complete official raw CSV. Compound map filters and street-level report drill-down are unavailable only for 2019; all other annual analysis remains complete.

How records are counted

Collision, party, and person tables describe different things. Party and person rows are never counted as additional crashes. When a filter uses those attributes, metrics count distinct collision IDs with at least one matching related record.

Geographic limitations

Coordinates are missing for a material portion of the archive. The dashboard does not invent coordinates from road names. All valid records remain in statewide and geographic-code summaries, while only records with usable coordinates appear as map cells or street-level points. A coverage percentage accompanies the map.

The opening density layer groups valid source coordinates into H3 resolution-7 cells, averaging about 5.2 square kilometers. Closer views progressively refine those cells through resolutions 8–10; at street zoom, each dot uses the latitude and longitude reported for that collision. Cell height and color are scaled within the visible layer for legibility—they show relative concentration, not physical height or roadway risk.

The cached 2026 opening layer is assigned from the reported coordinates directly. To keep historical and filtered statewide views responsive, live queries first aggregate coordinates into roughly two-kilometer source bins before assigning display cells; this may move a report into an adjacent overview cell but never changes the filtered totals. Zooming in requests finer source bins, and street-level dots use the unaggregated source coordinates.

Metric definitions

Reported collision
A distinct record in the CCRS crash table. The current source snapshot contains one row per collision ID, and none are marked deleted.
People injured or killed
Headline totals from the NumberInjured and NumberKilled fields on collision records. Person-level records are used for injury-severity and role analysis.
Injury-involved collision
A collision with at least one recorded injury or death in its crash-level totals.
Fatal or serious injury
A person-level record coded Fatal, SuspectSerious, or the legacy SevereInactive classification.
Hit-and-run
A collision with a misdemeanor or felony hit-and-run value recorded in CCRS.
At-fault party identified
A distinct collision with at least one related party whose source record is marked IsAtFault. This is a report field, not an independent legal conclusion.
Primary factor cited
A collision whose PrimaryCollisionFactorIsCited source field is marked true.
Report media flagged
A collision whose source record indicates photographs or digital media files. The public dataset provides the flag, not the underlying media.
Drug-recognition evaluation
A distinct collision with at least one related party record marked as having a drug-recognition evaluation conducted.
Recorded phone use
A distinct collision with at least one related party whose Special Information field indicates handheld or hands-free phone use. Different party categories may overlap within a collision.
Map coverage
The share of selected collision records with coordinates inside a broad California bounding region. Unmapped records remain in non-map totals. The map moves from statewide aggregation to exact reported coordinates as the user zooms.