How New Zealand actually works, the crash record
The crash record, what the map of hurt actually says
Every pin on the crashes pages is a line in the country's Crash Analysis System, the national register of crashes that police attended. It is the best road-safety dataset New Zealand has, and it is honest about being a partial one. Knowing what it counts is the difference between reading it and misreading it.
What it is
The Crash Analysis System, CAS, is maintained by NZ Transport Agency Waka Kotahi. It holds every crash a police officer recorded on or near a public road: where, when, who was hurt, what the diagram said, the factors the officer coded. It's a police operational record repurposed as a national dataset, which is exactly both its strength (it's official and consistent) and its limit (it only knows what was reported) — NZTA, the Crash Analysis System.
The severity ladder
CAS grades each crash by its worst outcome, one fatality makes the whole crash fatal:
- Fatal, someone died within 30 days from the crash's injuries. The national road toll is drawn from this tier.
- Serious, injuries severe enough for hospital admission or the medical equivalent, fractures, deep wounds, internal injuries. "Serious" in CAS is a clinical threshold, not a headline adjective.
- Minor, injury without hospital admission.
- Non-injury, damage only. The most numerous tier, and the one with the deepest under-reporting.
The tiers get quoted as DSI, deaths and serious injuries, the metric road-safety policy actually runs on — NZTA, road to zero.
The honest limits
- Reported only. A crash nobody calls the police about, most car-park prangs, many single-vehicle scrapes, the majority of bicycle falls, doesn't exist in CAS. Under-reporting is worst at the minor and non-injury tiers and for vulnerable users: cyclists and pedestrians who don't involve a car often never reach the record — NZTA, CAS.
- A count, not a rate. An intersection page that says "12 crashes in five years" says nothing about how much traffic it carries, a busy corner can hold more crashes and still be safer per crossing than an empty one. Exposure is the missing denominator; the site's pages are a record, not a risk model.
- Attribution is the officer's view. The coded factors, speed, alcohol, failed to give way, are the attending officer's assessment, useful at scale and fallible in the individual case.
Trends, hotspots and patterns. "This corner keeps producing serious crashes" is a CAS-supported claim. "This street is more dangerous than that one" usually isn't, not without exposure numbers. The intersections pages rank by count precisely because count is what the record honestly holds; read the rank as frequency, not danger-per-trip.
How to read a page
A crash entry is the record's own language: the severity first, then the movements, who hit what, in what direction, under what conditions. The year pages (/crashes/2024) are the city-scale view; the intersection pages are the same crashes asked "which corner". The queries worth asking: is the pattern concentrated (a fixable corner) or diffuse (a design-wide problem)? Is it daytime school traffic or 2 a.m.? The record holds both answers, the hour and the severity are printed on every entry.
Christchurch's crash map has its own signature, a flat grid with wide arterial roads and long intersection sightlines, where the hurting corners are the famous ones: the Moorhouse and Brougham corridors, the suburban roundabouts, the arterial T-junctions. The record also holds the east's quiet arithmetic, the suburbs the quake emptied report fewer crashes because there's fewer trips to have them.
Last reviewed 23 September 2026, sources: NZTA Crash Analysis System documentation, NZTA road-safety strategy materials, this site's crash record. Corrections: /corrections.
Questions, answered briefly
The national register of crashes that police attended, maintained by NZ Transport Agency Waka Kotahi, a police operational record repurposed as a dataset, which is both its strength and its limit: it is official and consistent, but it only knows what was reported. Every pin on the crashes pages is a CAS line.
No, a crash nobody calls the police about doesn't exist in CAS: most car-park prangs, many single-vehicle scrapes, the majority of bicycle falls. Under-reporting is worst at the minor and non-injury tiers and for cyclists and pedestrians who don't involve a car.
Injuries severe enough for hospital admission or the medical equivalent, fractures, deep wounds, internal injuries; a clinical threshold, not a headline adjective. A fatal crash is one where someone died within 30 days, and the tiers get quoted together as DSI, deaths and serious injuries, the metric road-safety policy runs on.
Usually not, the record holds counts, not rates, because the traffic-exposure denominator is missing; a busy corner can hold more crashes and still be safer per crossing than an empty one. What it does support is trends, hotspots and patterns: "this corner keeps producing serious crashes" is a CAS-backed claim, "this street is more dangerous than that one" usually isn't.