Between 2022 and 2023 Auckland lowered speed limits on hundreds of roads. In 2025 many were raised again. This page follows the crash record on 662 of those roads through all three periods.
662 roads with before / during / after recordsJanuary 2018 to April 2026Updated 11 June 2026
Adjusted total-crash estimatesOne of the four headline estimates is statistically clear: local streets during lower limits. Confidence intervals are shown below.
Corridors · during lower limit-7%relative to before
Local streets · during lower limit-17%relative to before
Corridors · after raised+5%relative to lower-limit period
Local streets · after raised+7%relative to lower-limit period
2026 is year-to-dateFull-month crash models end 30 April 2026May 2026 totals are still filling in and are excludedLater DSI records are kept separate
The bottom line
138 crashes prevented on local streets
while the 30 km/h limits applied, against what their own history predicted, the study’s one statistically clear result (95% CI 10 to 291 fewer, p = 0.034).
Best estimate: 33 fewer people killed or seriously injured
across all 662 study roads during the lower limits, roughly $66m in social cost at the official value of statistical life, but this estimate’s range still includes zero.
No detectable increase since limits went back up
41 people killed or seriously injured in the first 13 complete months; a corridor increase of 25% or more would already have shown. See each corridor →
01 / Adjusted results
Estimated changes across the three periods
While the lower limits applied, total crashes on local streets were an estimated 17% lower than before (95% CI -30% to -1%). Corridor crashes also sat below their pre-change level, though that interval includes no change. After the limits went back up, neither road group shows a clear change yet, the after period is short and the confidence intervals are wide. The local-street result keeps its size and direction under stricter checks, but some of those checks widen the interval enough to include zero.
Conditional Poisson phase contrasts use every informative strict common-geography road, adjust for traffic exposure, stable road differences, seasonality, COVID disruption and trend, and do not select roads based on post-change crashes.Sensitivity checks separately remove COVID-disrupted months, omit the traffic index, control Auckland-wide crash trends and retain only crashes rated high-confidence on both treatment geometries.
02 / Observed records
What happened on the study roads
Before any modelling, the raw record already shows the three-period shape: crash rates fell while the lower limits applied and have not returned to the old level in the short period since limits were raised. These charts are descriptive; the adjusted estimates above test how much of the pattern can be attributed to the speed-limit changes rather than to traffic, season or city-wide trends.
Observed rates use monthly Auckland traffic trends to make periods of different length more comparable.Roads are aligned to their own change dates. The five-month moving average retains visible variation, including the COVID-disrupted period.
03 / Roads later raised
Raised roads compared with roads that stayed lower
Two controlled checks. First, 26 raised corridors are compared with 52 corridors matched on pre-change records: a +18% change in total crashes after raising (95% CI -11% to +56%). Second, the 636 raised local streets are compared with 296 similar streets that kept 30 km/h: -5% for total crashes (95% CI -26% to +22%) and -26% for injury crashes (95% CI -54% to +21%). Neither comparison detects a change yet, and both total-crash pre-trend tests fail, so these are sensitivity results rather than primary estimates.
The corridor total-crash parallel-pretrend test is not satisfied (joint p = 0.064); the controlled estimate should therefore be interpreted cautiously.Raised local streets against 296 area-style streets that kept 30 km/h, with street and calendar-month fixed effects. The total-crash pre-trend test fails (p = 0.002); the injury comparison passes (p = 0.560) and is the more reliable of the two. The 1 July 2025 reversal date is assumed.
396raised-corridor crashes in the 12 pre months
400raised-corridor crashes in the 12 post months
134comparison-corridor crashes in the 12 pre months
115comparison-corridor crashes in the 12 post months
04 / Cross-checks
The corridor reversal tested several ways
Each method uses the records differently. Point estimates vary, and comparison-road estimates are limited by different pre-change trends.
Randomization inference, road bootstrap and comparison-road placebo tests also find no statistically clear corridor reversal effect. The synthetic estimate is less stable and is treated as a sensitivity check.The July 2025 local-road date is an assumed programme completion date. Moving it changes the estimated after-raising effect materially, so that result is preliminary.If raising the limits had increased total crashes by 25% or more, the corridor design would already be expected to detect it; a 15% increase becomes detectable around February 2027 for corridors and December 2028 for local streets. Poisson approximation without covariates; dates are indicative.
-24%corridor fake-date estimate one year before the real reduction
p = 0.020the corridor placebo detects a pre-existing change
26raised corridor groups in matched checks
52matched corridor groups that stayed lower
05 / Deaths and serious injuries
DSI outcomes across the three periods
The study includes crash severity and person-level outcomes. Suspected vehicle speeds add context, but they are investigator estimates rather than measured operating speeds. The local raised-period median uses only 7 records; 4 are coded 0 km/h.
Suspected speed is investigator-estimated and crash-conditional, not measured operating speed.Low agreement after changes may reflect partial implementation, CAS coding lag or remaining segment uncertainty. It limits causal interpretation.
322DSI crashes in the three-period study
361people killed or seriously injured
163pedestrian, cyclist or motorcyclist DSI
39DSI crashes after limits were raised
06 / Study roads
The study roads and crash matching
Official speed-change geometry and confirmed road fields define the study roads. The source inventory produces 791 continuous lower-then-raise sections; 662 have strict common geography and at least one crash in the complete-month study window.
Road-specific traffic counts remain uneven. Auckland-wide fixed-site traffic trends provide the main monthly exposure adjustment.
How the results were checked
COVIDDisrupted months are controlled and separately removed.
TrafficAT counts and Auckland-wide TMS trends are used as exposure checks.
BackgroundAn Auckland-wide monthly crash series checks common safety trends.
RobustnessNegative-binomial GEE, leave-one-road-out, randomization and placebo checks are included.
Local datesAll 636 local-road reversals currently use an assumed 1 July 2025 date.
Treatment inventory1,479 of 1,480 listed local-street reversal road names have a matching reduction record. Latham Avenue remains unmatched.
Area programmes442 area-style records have no reversal found; these are records, not yet validated unique areas.
07 / Technical files
Data, method and audit files
Every number on this page can be traced to a file below: model estimates, comparison panels, event-study coefficients and source audits.
How to read the results
The point estimates describe the direction and size of the estimated change. The confidence intervals describe the range still compatible with the available records. One of the four headline estimates is statistically clear: local streets during lower limits. Local-road after-raising estimates remain preliminary because the reversal date is assumed and no validated untreated-area comparison has yet been built.
Original pooled ITS estimates
Road group and outcome
During lower limit
After limit raised
CorridorsAll crashes
8% lower-21% to +8%p = 0.298
16% higher-5% to +42%p = 0.144
CorridorsInjury crashes
12% higher-15% to +48%p = 0.419
21% higher-16% to +76%p = 0.308
Local streetsAll crashes
8% lower-23% to +10%p = 0.370
100% lower-100% to +48516519441%p = 0.999
Local streetsInjury crashes
32% higher-6% to +85%p = 0.110
100% lower-100% to +48516519441%p = 0.999
Method in brief
Primary three-period estimates use conditional Poisson road-month models that condition out stable road differences and retain sparse roads without selecting on post-change crashes. The matched corridor comparison uses raised-segment geometry, road and calendar-month fixed effects, but its parallel-pretrend test fails. TMS fixed-site trends provide monthly traffic exposure. A true Empirical Bayes estimate still requires a calibrated Auckland safety-performance function and an untreated segment inventory.