STATS19 RESEARCH · PHASE 2

The parking estimate layer

Explore how housing, vehicle ownership and parking assumptions change England’s small-area off-street estimates.

Historical study stage · Parking estimates · Phase 2 — 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 2 · experimental estimates · Census 2021 areasParking scenarios are built. Local validation is outstanding.

Explore estimates from real housing and vehicle counts. The shares below depend on assumed parking capacity and behaviour; they are not observations of vehicles in garages.

33,755English LSOAs with housing and car/van counts joined

27 variantsCapacity, garage use, household allocation and housing-type mapping tested

Rates withheldMatched residential vehicle miles and independent parking validation still needed

Frozen 2026-10-10. These are 2021 household populations using 2024 national parking-provision evidence, not measured 2026 parking levels. Inspect the phase 1 collision-coding audit →

EXPLORE THE MEASUREMENT

How much parking might be off-street?

Start with a scenario, then inspect a small area. Every scenario is uncalibrated. “Reference” is a reproducible starting assumption, not a best estimate.

Capacity and use scenarioHousing and vehicle allocationCensus area name group

Assumptions in this view: 50% of garage-provision households with cars use one garage space; 1 additional off-street space at those homes; 1 space at homes with other off-street provision. “3 or more” vehicles is represented by 3.5.

Groups follow Census LSOA name prefixes for browsing; they are not a current authority crosswalk. Allocation alternatives preserve both Census margins. The housing types use national EHS probabilities in every area.

Find an LSOA in Bristol

Loading and checking the frozen area estimates…

Candidate deciles, before collision testing

Decile 1 has the lowest modelled off-street share; decile 10 has the highest. English LSOAs have equal rank weight. The displayed share pools modelled vehicles within each decile. Changing assumptions re-ranks all areas.

33.5%

38.5%

42.1%

45.2%

47.9%

50.2%

52.3%

54.2%

56.4%

D10

60.3%

National scenario: Reference assumptions · Independent housing and cars
Decile	LSOAs	Area share range	Pooled vehicle share	Collisions per 100,000 vehicle miles
1	3,376	23.5–36.3%	33.5%	Not calculated
2	3,375	36.3–40.4%	38.5%	Not calculated
3	3,376	40.4–43.7%	42.1%	Not calculated
4	3,375	43.7–46.5%	45.2%	Not calculated
5	3,376	46.5–49.1%	47.9%	Not calculated
6	3,375	49.1–51.3%	50.2%	Not calculated
7	3,376	51.3–53.3%	52.3%	Not calculated
8	3,375	53.3–55.2%	54.2%	Not calculated
9	3,376	55.2–57.8%	56.4%	Not calculated
10	3,375	57.8–74.1%	60.3%	Not calculated

Only 8 of 33,755 areas stay in their reference decile across all 27 variants. The median span is 6 decile steps. These are sensitivity-tested candidate groups, not evidence that garaging reduces accidents.

What does an independent benchmark tell us?

The National Travel Survey was kept out of the calculations. Its 2024 national overnight-parking shares provide a plausibility check, but cannot measure local model error.

Different dates and populations: synthetic 2021 households versus 2024 vehicle survey
Measure	Selected model	NTS 2024 England	Difference
Off-street vehicle share	49.6%	73.0%	-23.4 percentage points
Inside-garage vehicle share	11.8%	9.0%	+2.8 percentage points

The reference assumptions produce 49.6% off-street against the NTS national 73% benchmark: a shortfall of 23.4 percentage points. This is a material model mismatch, not a validated estimate. Changing capacity improves national agreement in some scenarios without establishing local accuracy.

NTS unweighted sample: 11,145 vehicles. Published percentages are rounded; street (25%) and other (3%) are separate categories. EHS household provision is 66.2%; the housing-weighted proxy is 68.7%. Provision and vehicle use have different denominators.

Reviewed; not comparable

Coventry · Earlsdon parking beat survey

Hourly beats, 06:00–24:00, weekday and Saturday, selected streets. Published percentages measure occupied spaces, not resident-vehicle parking shares. No matched LSOA denominator or full overnight coverage.

Identified; download blocked

Bristol · March 2025 parking survey package

Council describes daytime/night-time occupancy surveys. ZIP retrieval returned HTTP 403 in this run. Raw observations and exact LSOA/road matching were not verified.

Phase 2 validation remains open. No independent LSOA overnight survey has been matched to resident vehicles and road geometry. Local accuracy, calibration error and street vehicles per kilometre are therefore unavailable. National agreement would not validate local rankings.

What must pass before using these estimates in the accident model?

Collect dated overnight street counts, garage/driveway use and capacity in independent streets spanning the candidate deciles. Keep daytime visitors and commercial demand separate.
Match surveyed road segments to Census 2021 boundaries and residential centreline kilometres. Audit unmatched streets, cars parked outside their home area and population change.
Hold out whole local areas; report errors for both shares and street vehicles/km. Refit assumptions only on training observations, then freeze the estimator.
Verify STATS19 geography and eligible motor-collision roads; build traffic mileage for those same roads. Only then pool collisions and miles by decile.

Housing and car ownership must enter the later comparison model too. This derived score alone cannot prove an independent parking mechanism or causation.

Assumptions you can inspect

National EHS 2024 rates used in the reference mapping · provision, not use
Census housing category	Garage provision	Other off-street provision
Detached (including bungalows)	78.5%	20.6%
Semi-detached (including bungalows)	41.8%	43.2%
Terraced (including bungalows)	14.5%	34.2%
Purpose-built flat	6.0%	31.3%
Converted/shared house	2.9%	25.0%
Other converted building	2.9%	25.0%
Commercial building	2.9%	25.0%
Caravan/mobile structure	33.5%	32.6%

Commercial flats reuse the converted-flat rate; mobile homes reuse the all-dwelling rate. 1,601 areas have more than 5% of households in those two fallback categories. Sensitivity variants stress-test their provision from 0% to 100%.

Full estimator and decile rules

geography: England; Census 2021 LSOAs
reference: 2021 household counts with 2024 EHS provision rates; no annual backcast
householdReconciliation: Scale TS044 type counts proportionally to each TS045 household total. Preserve both published totals and their difference.
joint: Independent within each LSOA by default. Positive/negative rank couplings preserve both margins, concentrating more cars in higher/lower-provision types; they are stress tests, not joint Census observations.
capacity: For n vehicles, garage-use probability u and d extra spaces, garage-provision off-street capacity is u[min(n,1)+min(max(n−1,0),d)]+(1−u)min(n,d). Other off-street provision accommodates min(n,o). Apply these to synthetic household counts.
residual: Total modeled household vehicles minus off-street vehicles. This is street-or-other demand, not observed street parking. No per-km measure without matched residential centreline geometry.
deciles: Rank eligible English LSOAs by unrounded modeled vehicle off-street share, equal area weights. Cutpoints at ceil(N×d/10). Ties stay together in the lower decile; counts may differ. Re-rank separately for every scenario. Deciles are exposure candidates, not collision findings.
ranges: 27 variants = 3 capacity/use scenarios × 3 housing–car associations × 3 type mappings. Range is a sensitivity envelope, not a probability or confidence interval, and does not span every possible model.
mapping: House categories use EHS detached/semi/terraced rates, although Census includes bungalows. Purpose-built flats use the EHS population-weighted low/high-rise mix. Converted categories and commercial flats use converted-flat rates; mobile homes use all-dwelling rates. Alternative mappings substitute bungalow/flat extrema and 0–100% provision for commercial/mobile categories.
validation: National NTS benchmark held out from construction; numerical proximity is not validation. No aligned independent LSOA overnight vehicle-use validation, and no local error, correlation or holdout score is claimed.
interpretation: A housing/car-derived predictor cannot independently identify a parking effect beyond those ingredients. EHS rates describe provision, not capacity or use. Garage-use assumptions are not estimated from observed local behaviour.

National rates are transported to local housing types without local calibration. Capacity and garage use are illustrative assumptions. The 3+ car category has no observed mean here; testing 3–4 does not cover every possible household vehicle count.

Input reconciliation and quality checks

33,755 unique English LSOA keys in both tables; no unmatched housing or vehicle rows.
All within-table component counts sum to the published household totals.
22,669 LSOAs have different totals between the two Census tables; maximum absolute difference 9 households. Both totals remain downloadable.
Every synthetic joint preserves the reconciled housing and car-availability margins. Garage vehicles ≤ off-street vehicles ≤ total vehicles, in every scenario.
Unobserved parking, road length, collision counts and mileage remain null in JSON and blank in CSV.

Download and reproduce Phase 2

Frozen input bytes, model assumptions and all scenarios are available. The standard-library Python script runs offline from the repository and rebuilds every LSOA estimate and decile. No STATS19 outcomes were used to tune the estimates.

33,755 reference estimates · CSV All 27 variants · compressed CSV Methods, checks & benchmarks · JSON Rebuild script · Python Bristol · JSON

Modelled shares are percentages of household cars/vans under stated assumptions. Housing provision has a household denominator. No observed garaging percentages or collision rates are supplied. Full CSV metadata: data dictionary.

Sources · Open Government Licence v3.0 · retrieved 10 October 2026
Input	Observation period	Archived bytes
Census TS044	2021-03-21	Download snapshot SHA-256 `af7b512615ea9d27ad0b2cec715a6cd4b50832482e0f38ca640c906c352f91bd`
Census TS045	2021-03-21	Download snapshot SHA-256 `95593dcd53247d4458e97edd7454577a63e47fd43be957f0a2460e872efdc070`
EHS DA2201	2024	Download snapshot SHA-256 `d2bd11811c9dd8da846c0be8aa27db7382977263575318384c07810b78d5e2e9`
NTS0908	2024	Download snapshot SHA-256 `49f6044306df47428c8fd3a93053e8f01dae23425d6eb8c045d9ebc773a1a1fe`
EHS DA2203	2024	Download snapshot SHA-256 `cf7a27a9a5e0a5ba1b2fad109e06d285ba909914add67cb55c02b1f98791a1f8`

Census descriptions and disclosure-control guidance: Nomis bulk data. Housing and household populations differ from the EHS dwelling sample. EHS definitions and source tables. NTS parking-use benchmark.

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.