STATS19 RESEARCH · PHASE 3

Parking and collision frequency

Test the association between modelled parking shares and injury collisions per 100,000 vehicle miles—with the assumptions and geographic sensitivity in view.

Historical study stage · Descriptive frequencies · Phase 3 — Parking and collision risk · Adjusted analysis · Phase 4 →

Exploratory analysis · Exploratory adjusted analysis · England · 137 authorities · 2021–2025. Latest analysis date: 2026-10-10.

The context above describes the current study. This page retains the earlier evidence, period and limitations shown below.

Ask about the current released study →

Current release evidence summary (separate from this historical page)

Recorded decision: Exploratory incremental spatial prediction using an unvalidated proxy. No causal garage-use, driver-level or future-year validity claim.

Assessment: Leave one of nine English regions out, keeping all years for an authority together.

Evidence needed next: A prospectively frozen specification, uninspected outcomes and independent exposure validation before any pricing decision.

Full evaluation record and evidence hashes → · Study change history →

Study history and supporting evidence

Earlier analyses retain their original scope and results.

Phase 3 · exploratory authority analysis · all public roadsRates are calculated. A parking effect is not established.

Compare injury-collision frequency across modelled parking deciles using matched DfT authority mileage. The original residential-road LSOA test remains incomplete: neither local parking use nor residential mileage has been validated.

140 authoritiesThe same English authorities in all five years; 13 excluded

30,875 LSOAsParking estimates aggregated by modelled vehicles, using the ONS authority lookup

27 model variantsRetrospective, unadjusted comparisons; no held-out predictive test

Frozen 2026-10-10. Phase 2 estimates & validation gaps · Phase 1 coding audit

EXPLORE THE ASSOCIATION

Collision frequency by parking decile

Each decile groups authorities by modelled parking share. Collision counts and motor vehicle miles are then summed within each group. D1 is lowest parking; D10 is highest. These are authority deciles, distinct from the LSOA deciles in Phase 2.

Parking measureCapacity and use scenarioHousing and vehicle allocationOutcomePeriod

Selected assumptions: 50% of garage-provision households with cars use a garage space; 1 additional off-street spaces there; 1 spaces at homes with other off-street provision; 3.5 vehicles for the Census “3+” category. Reference housing-type mapping; all mappings are tested below. These are assumptions, not measured behaviour.

0.13×D10 / D1 rate · 2021–2025 pooled
95% authority bootstrap interval: 0.09–0.18

0.11–5.63×D10 / D1 range across all 27 model variants
Direction reverses across assumptions.

0.71× outside LondonPooled 2021–2025, re-ranked without London
95% authority bootstrap: 0.40–1.00

The direction is not robust to parking assumptions. Some tested models associate more off-street parking with lower frequency; others associate it with higher frequency. The selected D1 contains 13 London authorities; D10 contains 0. Road mix, urbanity, housing and car ownership are not controlled. Excluding London is a sensitivity check, not an adjusted model.

The numerator includes all reported injury collisions, including those with no motor vehicle. The denominator is all motor vehicle travel within each authority, including through traffic. This is an area event frequency measure, not an individual driver’s collision probability. Damage-only crashes are absent.

All reported injury collisions · 2021–2025 pooled. Collision events per 100,000 motor vehicle miles. Whiskers: 95% exact Poisson intervals, conditional on the estimated miles. These narrow intervals do not capture differences between authorities or uncertainty in parking estimates.

All reported injury collisions · 2021–2025 pooled · same 140 authorities
Decile	Modelled share range	Authorities	Collisions	Million vehicle miles	Per 100,000 miles	95% Poisson interval
D1	31.0–38.8%	14	52,797	26,187	0.2016	0.1999–0.2033
D2	40.3–43.7%	14	42,242	49,204	0.0859	0.0850–0.0867
D3	43.7–46.3%	14	37,274	82,465	0.0452	0.0447–0.0457
D4	46.3–48.2%	14	58,745	131,982	0.0445	0.0442–0.0449
D5	48.3–49.4%	14	53,257	160,773	0.0331	0.0328–0.0334
D6	49.5–50.6%	14	50,032	195,618	0.0256	0.0254–0.0258
D7	50.6–51.3%	14	49,310	207,690	0.0237	0.0235–0.0240
D8	51.3–52.4%	14	28,477	102,775	0.0277	0.0274–0.0280
D9	52.6–53.5%	14	33,243	134,802	0.0247	0.0244–0.0249
D10	53.5–56.7%	14	41,582	160,640	0.0259	0.0256–0.0261
Pooled across all deciles			446,959	1,252,137	0.0357	Not an average of decile rates

Miles display rounded to millions; calculations use source miles. Share ranges show the minimum and maximum authority shares, rounded for display. Ties stay together, so bins need not contain equal numbers. The bootstrap holds decile membership fixed and resamples whole authorities; it does not include model measurement error.

Does the pattern survive different assumptions?

Compare the selected outcome and period across nine capacity and allocation choices under the reference housing mapping. Every row rebuilds the parking ranking on the same cohort. The full 27-variant range additionally changes the housing mapping.

All reported injury collisions · 2021–2025 pooled · reference housing mapping
Capacity and use	Housing–vehicle allocation	D10 / D1	95% authority bootstrap interval
Ample capacity	Independent housing and cars	0.11×	0.09–0.14
Ample capacity	More cars in lower-provision housing	0.11×	0.09–0.14
Ample capacity	More cars in higher-provision housing	0.12×	0.09–0.16
Constrained capacity	Independent housing and cars	1.85×	1.28–2.58
Constrained capacity	More cars in lower-provision housing	1.24×	0.84–1.71
Constrained capacity	More cars in higher-provision housing	3.77×	2.46–6.16
Reference assumptions · selected	Independent housing and cars	0.13×	0.09–0.18
Reference assumptions	More cars in lower-provision housing	0.11×	0.09–0.14
Reference assumptions	More cars in higher-provision housing	0.20×	0.14–0.32

All 27 model variants for the selected outcome and period

Model sensitivity · not a confidence distribution
Housing mapping	Capacity	Allocation	D10 / D1
reference	ample	independent	0.11×
reference	ample	negative	0.11×
reference	ample	positive	0.12×
reference	constrained	independent	1.85×
reference	constrained	negative	1.24×
reference	constrained	positive	3.77×
reference	reference	independent	0.13×
reference	reference	negative	0.11×
reference	reference	positive	0.20×
higher-provision	ample	independent	0.11×
higher-provision	ample	negative	0.11×
higher-provision	ample	positive	0.12×
higher-provision	constrained	independent	1.98×
higher-provision	constrained	negative	1.35×
higher-provision	constrained	positive	4.67×
higher-provision	reference	independent	0.14×
higher-provision	reference	negative	0.13×
higher-provision	reference	positive	0.69×
lower-provision	ample	independent	0.11×
lower-provision	ample	negative	0.11×
lower-provision	ample	positive	0.13×
lower-provision	constrained	independent	5.01×
lower-provision	constrained	negative	1.62×
lower-provision	constrained	positive	5.63×
lower-provision	reference	independent	0.15×
lower-provision	reference	negative	0.12×
lower-provision	reference	positive	0.21×

Same deciles across years

The selected parking estimates and authority membership stay fixed. This retrospective comparison is not predictive validation: the parking model combines Census 2021 households with EHS 2024 provision rates. The 2021 outcome is also affected by the pandemic.

All reported injury collisions · selected parking assumptions
Year	D1 rate	D10 rate	D10 / D1	95% authority bootstrap interval
2021	0.2193	0.0274	0.12×	0.09–0.17
2022	0.2115	0.0262	0.12×	0.09–0.17
2023	0.2092	0.0255	0.12×	0.09–0.17
2024	0.1836	0.0253	0.14×	0.10–0.19
2025	0.1853	0.0252	0.14×	0.10–0.18

Rates per 100,000 motor vehicle miles. Repeated years are not independent replication of the parking predictor.

Which authorities are in each decile?

Inspect a decile

D1 · selected parking measure and assumptions · frozen for 2021–2025
Authority	Code	Region	Modelled vehicle share	LSOAs
City of London	E09000001	London	30.98%	6
Kensington and Chelsea	E09000020	London	31.49%	101
Westminster	E09000033	London	31.89%	123
Camden	E09000007	London	32.92%	130
Islington	E09000019	London	33.45%	126
Tower Hamlets	E09000030	London	34.14%	169
Hammersmith and Fulham	E09000013	London	34.38%	115
Hackney	E09000012	London	35.00%	149
Lambeth	E09000022	London	36.06%	181
Wandsworth	E09000032	London	36.22%	186
Southwark	E09000028	London	36.47%	173
Haringey	E09000014	London	37.76%	147
Portsmouth	E06000044	South East	38.56%	125
Lewisham	E09000023	London	38.82%	175

What is included—and what is still missing?

700 authority-years pass the fixed-cohort checks. Every included collision count reconciles with the Phase 1 audit and official authority total. Traffic and parking geography join by exact code. All roads are included; no residential-road or LSOA mileage is manufactured.

Coverage · all reported injury collisions
Year	Included collisions	Excluded collisions	Included million vehicle miles
2021	88,234	5,639	228,844
2022	93,000	5,546	248,510
2023	90,861	5,903	253,990
2024	87,280	6,567	257,908
2025	87,584	6,834	262,886

13 excluded authorities and exact reasons

Failure in any year excludes an authority from all five years. Code changes are not repaired by name matching.

Excluded from every decile and period
Authority	Failing years	Reasons	2021–2025 collisions
Cumberland E06000063	2021, 2022	No matching traffic authority code and year; mileage rate withheld; No positive exact-code DfT authority/year mileage; Upstream collision/mileage reconciliation did not pass	2,045
Westmorland and Furness E06000064	2021, 2022	No matching traffic authority code and year; mileage rate withheld; No positive exact-code DfT authority/year mileage; Upstream collision/mileage reconciliation did not pass	1,990
Barnsley E08000038	2021, 2022, 2023, 2024, 2025	No exact UTLA23 lookup code; no name matching or boundary apportionment; No matching traffic authority code and year; mileage rate withheld; No positive exact-code DfT authority/year mileage; Upstream collision/mileage reconciliation did not pass	1,800
Sheffield E08000039	2021, 2022, 2023, 2024, 2025	No exact UTLA23 lookup code; no name matching or boundary apportionment; No matching traffic authority code and year; mileage rate withheld; No positive exact-code DfT authority/year mileage; Upstream collision/mileage reconciliation did not pass	3,953
Staffordshire E10000028	2022, 2023	DfT warning: Avon and Somerset / Staffordshire reporting may be incomplete	4,178
Stoke-on-Trent E06000021	2022, 2023	DfT warning: Avon and Somerset / Staffordshire reporting may be incomplete	1,184
Buckinghamshire E06000060	2021, 2022, 2023, 2024, 2025	Official comparison rate unavailable; independent reconciliation not possible	3,312
Bath and North East Somerset E06000022	2022, 2023	DfT warning: Avon and Somerset / Staffordshire reporting may be incomplete	832
Bristol, City of E06000023	2022, 2023	DfT warning: Avon and Somerset / Staffordshire reporting may be incomplete	4,110
Isles of Scilly E06000053	2022, 2023, 2025	Traffic denominator differs by over 0.5% from the official rate; mileage rate withheld; Upstream collision/mileage reconciliation did not pass	9
North Somerset E06000024	2022, 2023	DfT warning: Avon and Somerset / Staffordshire reporting may be incomplete	1,293
Somerset E06000066	2022, 2023	DfT warning: Avon and Somerset / Staffordshire reporting may be incomplete	3,984
South Gloucestershire E06000025	2022, 2023	DfT warning: Avon and Somerset / Staffordshire reporting may be incomplete	1,799

The parking predictor is unvalidated. Phase 2’s reference national off-street estimate was 49.6%, compared with the NTS benchmark of 73.0%. This mismatch, the assumption-sensitive decile ranking and missing local parking observations limit interpretation. Inspect the validation evidence.

Next analysis gate

Validate local parking use and match residential-road exposure before the original LSOA mechanism test. Phase 4 should test whether parking adds predictive information after urbanity, road mix, deprivation, housing and car ownership, using whole-area holdouts and a genuinely later outcome period. These Phase 3 results do not establish that incremental value.

Full methods, uncertainty and scope

scope: Exploratory English authority analysis on ALL public roads. Numerator: all police-reported injury collisions, including non-motor-only collisions; denominator: DfT all-motor vehicle miles. This is an area-level event frequency measure, not per-driver collision probability or a matched residential/motor-only cohort.
geography: LSOA21 estimates are assigned using the ONS UTLA23 best-fit lookup. Exact authority codes join to current-highway-authority collision counts and year-specific traffic. No LSOA mileage is inferred. Barnsley and Sheffield code differences remain excluded rather than name-matched.
cohort: Balanced authority cohort with valid parking lookup and collision/mileage reconciliation in every year 2021–2025. An authority failing any year is excluded from every year, avoiding changing composition. No outcome magnitude is used to select deciles.
parking: For each authority, sum modelled off-street (or inside-garage) vehicles over LSOAs and divide by total modelled vehicles. Reuse all 27 frozen Phase 2 variants at their published numeric precision. These parking shares are unvalidated proxies.
deciles: Rank the fixed included authorities by parking share with equal authority weights. Cutpoints at ceil(N*d/10); ties stay together in the lower decile. Freeze the same membership across years within each scenario. These are AUTHORITY deciles, not the Phase 2 LSOA deciles.
rates: 100,000 * summed collisions / summed motor vehicle miles on the same included authority-years. Never average authority rates. Unknown parking status remains in total collisions and mileage; recorded parked involvement is a lower count, with an upper sensitivity count assuming every unresolved case involved a parked vehicle.
uncertainty: Per-decile 95% exact Poisson (Garwood) intervals condition on fixed traffic estimates and ignore spatial clustering/overdispersion. D10/D1 also has a 95% percentile authority bootstrap: 2,000 samples within the two fixed bins, with all years for an authority resampled together. Neither interval covers parking measurement error, traffic estimation error or systematic under-reporting.
contrast: D10/D1 = highest-parking decile rate divided by lowest-parking decile rate. Poisson ratio interval uses log-rate approximation. No hypothesis-selection correction or causal interpretation. The scenario envelope is not a confidence interval.
london: Additional pooled sensitivity removes London and re-ranks the remainder. This checks geographic concentration, not confounding adjustment. Urbanity, road mix, deprivation, housing and vehicle ownership still require Phase 4 controls.
time: Retrospective descriptive comparison: 2024 EHS rates are applied to 2021 household populations, then compared with 2021–2025 outcomes. Annual consistency is not out-of-time predictive validation. 2021 is pandemic-affected.
limits: STATS19 excludes damage-only collisions. A parked vehicle being involved does not establish contact, fault or causation. Missing involvement coding varies by authority/year. Residential LSOA frequency and independently observed garage-use effects remain untested.

Download and reproduce Phase 3

The fixed input manifest records source paths and SHA-256 checksums. The Python builder runs from the repository root with NumPy 2.3.5 and SciPy 1.17.0; it preserves exclusions, recomputes every decile and uses a fixed seed for 2,000 authority bootstrap samples.

All decile rates · CSV Authority-year panel · CSV Every decile assignment · CSV Scenario contrasts · CSV Results & methods · JSON Frozen input manifest Rebuild script · Python Data dictionary Reproduction guide

Source register · archived inputs available under the Open Government Licence v3.0
Source	Period	Input snapshot
Phase 1 collision and parked-involvement audit	2021–2025	Download
Phase 2 frozen parking scenarios	2021 households; 2024 provision	Download
RAS0403: collisions and casualties by local authority	2016–2025	Download
Local authority motor vehicle mileage	2016–2025	Download
ONS LSOA21 → UTLA23 best-fit lookup	LSOA 2021 / authority 2023	Download

Frozen authority metadata and upstream quality decisions. Source rate reconciliation tolerance: 0.5%. DfT mileage is estimated; neither Poisson nor bootstrap intervals include its estimation error. Small-area traffic estimates require particular care: DfT methodology.

Johan Strydom

Loading approved evidence

Retrieving and verifying the requested analytical release. Result data loads before optional map boundaries and source checks.

Your filters and release stay in the address. Navigation remains available.

STATS19 RESEARCH · PHASE 3

Parking and collision frequency

Test the association between modelled parking shares and injury collisions per 100,000 vehicle miles—with the assumptions and geographic sensitivity in view.

Historical study stage · Descriptive frequencies · Phase 3 — Parking and collision risk · Adjusted analysis · Phase 4 →

Exploratory analysis · Exploratory adjusted analysis · England · 137 authorities · 2021–2025. Latest analysis date: 2026-10-10.

The context above describes the current study. This page retains the earlier evidence, period and limitations shown below.

Ask about the current released study →

Current release evidence summary (separate from this historical page)

Recorded decision: Exploratory incremental spatial prediction using an unvalidated proxy. No causal garage-use, driver-level or future-year validity claim.

Assessment: Leave one of nine English regions out, keeping all years for an authority together.

Evidence needed next: A prospectively frozen specification, uninspected outcomes and independent exposure validation before any pricing decision.

Full evaluation record and evidence hashes → · Study change history →

Study history and supporting evidence

Earlier analyses retain their original scope and results.

Phase 3 · exploratory authority analysis · all public roadsRates are calculated. A parking effect is not established.

140 authoritiesThe same English authorities in all five years; 13 excluded

30,875 LSOAsParking estimates aggregated by modelled vehicles, using the ONS authority lookup

27 model variantsRetrospective, unadjusted comparisons; no held-out predictive test

Frozen 2026-10-10. Phase 2 estimates & validation gaps · Phase 1 coding audit

EXPLORE THE ASSOCIATION

Collision frequency by parking decile

Parking measureCapacity and use scenarioHousing and vehicle allocationOutcomePeriod

0.13×D10 / D1 rate · 2021–2025 pooled
95% authority bootstrap interval: 0.09–0.18

0.11–5.63×D10 / D1 range across all 27 model variants
Direction reverses across assumptions.

0.71× outside LondonPooled 2021–2025, re-ranked without London
95% authority bootstrap: 0.40–1.00

All reported injury collisions · 2021–2025 pooled · same 140 authorities
Decile	Modelled share range	Authorities	Collisions	Million vehicle miles	Per 100,000 miles	95% Poisson interval
D1	31.0–38.8%	14	52,797	26,187	0.2016	0.1999–0.2033
D2	40.3–43.7%	14	42,242	49,204	0.0859	0.0850–0.0867
D3	43.7–46.3%	14	37,274	82,465	0.0452	0.0447–0.0457
D4	46.3–48.2%	14	58,745	131,982	0.0445	0.0442–0.0449
D5	48.3–49.4%	14	53,257	160,773	0.0331	0.0328–0.0334
D6	49.5–50.6%	14	50,032	195,618	0.0256	0.0254–0.0258
D7	50.6–51.3%	14	49,310	207,690	0.0237	0.0235–0.0240
D8	51.3–52.4%	14	28,477	102,775	0.0277	0.0274–0.0280
D9	52.6–53.5%	14	33,243	134,802	0.0247	0.0244–0.0249
D10	53.5–56.7%	14	41,582	160,640	0.0259	0.0256–0.0261
Pooled across all deciles			446,959	1,252,137	0.0357	Not an average of decile rates

Does the pattern survive different assumptions?

All reported injury collisions · 2021–2025 pooled · reference housing mapping
Capacity and use	Housing–vehicle allocation	D10 / D1	95% authority bootstrap interval
Ample capacity	Independent housing and cars	0.11×	0.09–0.14
Ample capacity	More cars in lower-provision housing	0.11×	0.09–0.14
Ample capacity	More cars in higher-provision housing	0.12×	0.09–0.16
Constrained capacity	Independent housing and cars	1.85×	1.28–2.58
Constrained capacity	More cars in lower-provision housing	1.24×	0.84–1.71
Constrained capacity	More cars in higher-provision housing	3.77×	2.46–6.16
Reference assumptions · selected	Independent housing and cars	0.13×	0.09–0.18
Reference assumptions	More cars in lower-provision housing	0.11×	0.09–0.14
Reference assumptions	More cars in higher-provision housing	0.20×	0.14–0.32

All 27 model variants for the selected outcome and period

Model sensitivity · not a confidence distribution
Housing mapping	Capacity	Allocation	D10 / D1
reference	ample	independent	0.11×
reference	ample	negative	0.11×
reference	ample	positive	0.12×
reference	constrained	independent	1.85×
reference	constrained	negative	1.24×
reference	constrained	positive	3.77×
reference	reference	independent	0.13×
reference	reference	negative	0.11×
reference	reference	positive	0.20×
higher-provision	ample	independent	0.11×
higher-provision	ample	negative	0.11×
higher-provision	ample	positive	0.12×
higher-provision	constrained	independent	1.98×
higher-provision	constrained	negative	1.35×
higher-provision	constrained	positive	4.67×
higher-provision	reference	independent	0.14×
higher-provision	reference	negative	0.13×
higher-provision	reference	positive	0.69×
lower-provision	ample	independent	0.11×
lower-provision	ample	negative	0.11×
lower-provision	ample	positive	0.13×
lower-provision	constrained	independent	5.01×
lower-provision	constrained	negative	1.62×
lower-provision	constrained	positive	5.63×
lower-provision	reference	independent	0.15×
lower-provision	reference	negative	0.12×
lower-provision	reference	positive	0.21×

Same deciles across years

All reported injury collisions · selected parking assumptions
Year	D1 rate	D10 rate	D10 / D1	95% authority bootstrap interval
2021	0.2193	0.0274	0.12×	0.09–0.17
2022	0.2115	0.0262	0.12×	0.09–0.17
2023	0.2092	0.0255	0.12×	0.09–0.17
2024	0.1836	0.0253	0.14×	0.10–0.19
2025	0.1853	0.0252	0.14×	0.10–0.18

Rates per 100,000 motor vehicle miles. Repeated years are not independent replication of the parking predictor.

Which authorities are in each decile?

Inspect a decile

D1 · selected parking measure and assumptions · frozen for 2021–2025
Authority	Code	Region	Modelled vehicle share	LSOAs
City of London	E09000001	London	30.98%	6
Kensington and Chelsea	E09000020	London	31.49%	101
Westminster	E09000033	London	31.89%	123
Camden	E09000007	London	32.92%	130
Islington	E09000019	London	33.45%	126
Tower Hamlets	E09000030	London	34.14%	169
Hammersmith and Fulham	E09000013	London	34.38%	115
Hackney	E09000012	London	35.00%	149
Lambeth	E09000022	London	36.06%	181
Wandsworth	E09000032	London	36.22%	186
Southwark	E09000028	London	36.47%	173
Haringey	E09000014	London	37.76%	147
Portsmouth	E06000044	South East	38.56%	125
Lewisham	E09000023	London	38.82%	175

What is included—and what is still missing?

Coverage · all reported injury collisions
Year	Included collisions	Excluded collisions	Included million vehicle miles
2021	88,234	5,639	228,844
2022	93,000	5,546	248,510
2023	90,861	5,903	253,990
2024	87,280	6,567	257,908
2025	87,584	6,834	262,886

13 excluded authorities and exact reasons

Failure in any year excludes an authority from all five years. Code changes are not repaired by name matching.

Excluded from every decile and period
Authority	Failing years	Reasons	2021–2025 collisions
Cumberland E06000063	2021, 2022	No matching traffic authority code and year; mileage rate withheld; No positive exact-code DfT authority/year mileage; Upstream collision/mileage reconciliation did not pass	2,045
Westmorland and Furness E06000064	2021, 2022	No matching traffic authority code and year; mileage rate withheld; No positive exact-code DfT authority/year mileage; Upstream collision/mileage reconciliation did not pass	1,990
Barnsley E08000038	2021, 2022, 2023, 2024, 2025	No exact UTLA23 lookup code; no name matching or boundary apportionment; No matching traffic authority code and year; mileage rate withheld; No positive exact-code DfT authority/year mileage; Upstream collision/mileage reconciliation did not pass	1,800
Sheffield E08000039	2021, 2022, 2023, 2024, 2025	No exact UTLA23 lookup code; no name matching or boundary apportionment; No matching traffic authority code and year; mileage rate withheld; No positive exact-code DfT authority/year mileage; Upstream collision/mileage reconciliation did not pass	3,953
Staffordshire E10000028	2022, 2023	DfT warning: Avon and Somerset / Staffordshire reporting may be incomplete	4,178
Stoke-on-Trent E06000021	2022, 2023	DfT warning: Avon and Somerset / Staffordshire reporting may be incomplete	1,184
Buckinghamshire E06000060	2021, 2022, 2023, 2024, 2025	Official comparison rate unavailable; independent reconciliation not possible	3,312
Bath and North East Somerset E06000022	2022, 2023	DfT warning: Avon and Somerset / Staffordshire reporting may be incomplete	832
Bristol, City of E06000023	2022, 2023	DfT warning: Avon and Somerset / Staffordshire reporting may be incomplete	4,110
Isles of Scilly E06000053	2022, 2023, 2025	Traffic denominator differs by over 0.5% from the official rate; mileage rate withheld; Upstream collision/mileage reconciliation did not pass	9
North Somerset E06000024	2022, 2023	DfT warning: Avon and Somerset / Staffordshire reporting may be incomplete	1,293
Somerset E06000066	2022, 2023	DfT warning: Avon and Somerset / Staffordshire reporting may be incomplete	3,984
South Gloucestershire E06000025	2022, 2023	DfT warning: Avon and Somerset / Staffordshire reporting may be incomplete	1,799

Next analysis gate

Full methods, uncertainty and scope

scope: Exploratory English authority analysis on ALL public roads. Numerator: all police-reported injury collisions, including non-motor-only collisions; denominator: DfT all-motor vehicle miles. This is an area-level event frequency measure, not per-driver collision probability or a matched residential/motor-only cohort.
geography: LSOA21 estimates are assigned using the ONS UTLA23 best-fit lookup. Exact authority codes join to current-highway-authority collision counts and year-specific traffic. No LSOA mileage is inferred. Barnsley and Sheffield code differences remain excluded rather than name-matched.
cohort: Balanced authority cohort with valid parking lookup and collision/mileage reconciliation in every year 2021–2025. An authority failing any year is excluded from every year, avoiding changing composition. No outcome magnitude is used to select deciles.
parking: For each authority, sum modelled off-street (or inside-garage) vehicles over LSOAs and divide by total modelled vehicles. Reuse all 27 frozen Phase 2 variants at their published numeric precision. These parking shares are unvalidated proxies.
deciles: Rank the fixed included authorities by parking share with equal authority weights. Cutpoints at ceil(N*d/10); ties stay together in the lower decile. Freeze the same membership across years within each scenario. These are AUTHORITY deciles, not the Phase 2 LSOA deciles.
rates: 100,000 * summed collisions / summed motor vehicle miles on the same included authority-years. Never average authority rates. Unknown parking status remains in total collisions and mileage; recorded parked involvement is a lower count, with an upper sensitivity count assuming every unresolved case involved a parked vehicle.
uncertainty: Per-decile 95% exact Poisson (Garwood) intervals condition on fixed traffic estimates and ignore spatial clustering/overdispersion. D10/D1 also has a 95% percentile authority bootstrap: 2,000 samples within the two fixed bins, with all years for an authority resampled together. Neither interval covers parking measurement error, traffic estimation error or systematic under-reporting.
contrast: D10/D1 = highest-parking decile rate divided by lowest-parking decile rate. Poisson ratio interval uses log-rate approximation. No hypothesis-selection correction or causal interpretation. The scenario envelope is not a confidence interval.
london: Additional pooled sensitivity removes London and re-ranks the remainder. This checks geographic concentration, not confounding adjustment. Urbanity, road mix, deprivation, housing and vehicle ownership still require Phase 4 controls.
time: Retrospective descriptive comparison: 2024 EHS rates are applied to 2021 household populations, then compared with 2021–2025 outcomes. Annual consistency is not out-of-time predictive validation. 2021 is pandemic-affected.
limits: STATS19 excludes damage-only collisions. A parked vehicle being involved does not establish contact, fault or causation. Missing involvement coding varies by authority/year. Residential LSOA frequency and independently observed garage-use effects remain untested.

Download and reproduce Phase 3

Source register · archived inputs available under the Open Government Licence v3.0
Source	Period	Input snapshot
Phase 1 collision and parked-involvement audit	2021–2025	Download
Phase 2 frozen parking scenarios	2021 households; 2024 provision	Download
RAS0403: collisions and casualties by local authority	2016–2025	Download
Local authority motor vehicle mileage	2016–2025	Download
ONS LSOA21 → UTLA23 best-fit lookup	LSOA 2021 / authority 2023	Download