23 April 2023 test
RR 0.62
3 observed · 4.81 control mean · empirical p 0.784
No positive excess on the pre-specified primary endpoint.
An SCC Nexus research observatory · UK
What happens after the alarm?
Reproducible analysis of whether population-scale mobile Emergency Alerts are followed by measurable unintended effects in road safety, health, safeguarding and public disruption.
Study
Active
Primary endpoint
0–<15 min collisions
2026 wildfire case
Awaiting outcome data
Public update
31 Aug 2026
Historical evidence
The primary result is currently negative.
The pre-specified endpoint is police-reported personal-injury collisions from the archived alert time to 15 minutes later. Association is not causation.
7 September 2025 test
RR 1.26
7 observed · 5.56 control mean · empirical p 0.197
The primary endpoint is not statistically unusual in the matched-control analysis.
2025 first five minutes
RR 1.74
5 observed · 2.88 control mean · empirical p 0.057
This was a pre-specified secondary sensitivity window, not a post-hoc replacement for the 0–<15 minute primary endpoint. Because several secondary windows are examined, the value is treated as exploratory, non-confirmatory and subject to multiplicity. It must not be promoted as evidence that alerts increase collisions.
At-a-glance graph
Observed collisions versus matched controls.
Counts refer to the pre-specified 0–<15 minute window. The graphic is descriptive; the matched-control empirical test remains the inferential result.
2023 observed3.00
2025 observed7.00
| Alert | Window | Observed | Control mean | Rate ratio | Empirical p | Interpretation |
|---|---|---|---|---|---|---|
| 23 Apr 2023 | 0–<15 min | 3 | 4.8125 | 0.623 | 0.784 | No positive excess |
| 7 Sep 2025 | 0–<15 min | 7 | 5.5625 | 1.258 | 0.197 | Not statistically unusual |
| 7 Sep 2025 | 0–<5 min | 5 | 2.875 | 1.739 | 0.057 | Pre-specified secondary; exploratory only |
Secondary windows (0–<5, 30 and 60 minutes) are sensitivity analyses. Their p-values are not treated as independent confirmatory tests. See Methodology for the analysis hierarchy and multiplicity guardrail.
The rule that governs this Observatory
Statistical signal ≠ causation. Discovery, sensitivity tests and public evidence are deliberately separated. A result only becomes public after provenance and publication checks.
Current historical coverage
205,757
DfT STATS19 personal-injury collision records ingested across the 2023 and 2025 historical baselines.
Alert cases
From planned tests to operational warnings.
Operational alerts are especially informative because recipients may have less warning than during nationally publicised tests.
23 Apr 202314:59 BST
National Emergency Alert testGreat Britain · historical road-safety analysis complete
Analysed7 Sep 202515:00 BST
National Emergency Alert testGreat Britain · historical road-safety analysis complete
Analysed14 Aug 202618:57 BST
Operational wildfire alert — WalesWelsh-language broadcast; retained as a distinct timestamped case.
Awaiting data14 Aug 202619:01 BST
Operational wildfire alert — England & WalesMost extensive active-incident use to date; suitable collision outcome data not yet available.
Awaiting dataEvidence lanes
Separate outcomes. No synthetic “harm score”.
Each domain has its own evidence quality and timing constraints. Missing data is shown as missing rather than silently inferred.
Road collisionsHistorical analysis active
Traffic behaviourSource development
Emergency medicineAwaiting suitable aggregate data
Domestic abuse & safeguardingAwaiting timestamped aggregate data
Transport disruptionPlanned
Live-event disruptionPlanned
Latest context
14 August 2026 wildfire alert
The UK Government said the alert accompanied a wider wildfire-response package including additional fire-service support, military assistance and a temporary disposable-barbecue ban. The Observatory records this as context, not as evidence that the alert caused or prevented any measured outcome.
Important limitation
STATS19 does not measure every road incident.
It covers police-reported personal-injury collisions. Damage-only bumps and near misses may be absent, so a null STATS19 finding does not prove that the alarm caused no driving disturbance.