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.
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.
ORIGINAL PHASE 5 · ENGLAND & WALES · FROZEN NATIONAL DECILES
Each available feature is tested against the geographic and road baseline. Predictive improvement, regional persistence and mileage sensitivity determine the research shortlist. Null and reversed results stay visible.
This page preserves the Phase 5 results for the original 38 banded features. F7 collision screening and the expanded ranking are now available for the two admitted additions; the other nine remain explicitly untested.
IMPLEMENTED · RESEARCH SHORTLIST
All 35,672 England and Wales LSOAs retain their original national bands. Of 38 tested features, 28 are new candidates and 10 test the form of variables already in the baseline. Six diagnostic-only features and 18 deferred candidates remain in the register without invented deciles.
This is retrospective development evidence. The mileage allocation remains provisional, the baseline has calibration limitations, and 2024/25 outcomes have already been inspected. A research shortlist does not prove the paper’s hypothesis or establish value against insurer claims.
| Feature | Test family | 2023 deviance gain | Geographic deviance gain | Positive holdouts |
|---|---|---|---|---|
| Social rented households | novel candidate | 1.07% | 1.94% | 8/10 |
| Private rented households | novel candidate | 3.10% | 2.10% | 9/10 |
| Owner occupied households | novel candidate | 4.10% | 4.61% | 9/10 |
| Overcrowded households | novel candidate | 3.61% | 10.28% | 10/10 |
| Unemployment among economically active adults | novel candidate | 3.03% | 2.66% | 9/10 |
| Flats and converted/commercial accommodation | novel candidate | 3.71% | 8.73% | 10/10 |
| Households without a car or van | novel candidate | 7.05% | 7.49% | 8/10 |
| Cars/vans per household · lower bound | novel candidate | 6.66% | 7.21% | 8/10 |
| Households with 2+ cars or vans | novel candidate | 6.28% | 7.26% | 8/10 |
| Households with 3+ cars or vans | novel candidate | 4.29% | 4.56% | 8/10 |
| Car/van driver commuting share | novel candidate | 5.44% | 13.64% | 10/10 |
| Public transport connectivity | novel candidate | 5.64% | 12.99% | 9/10 |
| Walking connectivity | novel candidate | 6.96% | 8.66% | 10/10 |
| Cycling connectivity | novel candidate | 4.45% | 7.87% | 8/10 |
| Driving connectivity | novel candidate | 2.63% | 4.02% | 8/10 |
Fit 2022 and predict 2023, then hold each English region or Wales out across 2022–23. Every feature is compared with a baseline refitted on its own eligible rows. Zero-collision observations remain in the count models, with log(miles / 100,000) as the offset.
Categorical national bands are the primary test. Linear values and restricted cubic splines provide secondary comparisons. False-discovery rates are controlled separately across the 28 new candidates and 10 baseline-form diagnostics.
Adjusted band effects are refitted by development year, region and urban/rural context. Alternative mileage allocations use the same eligible rows. A separate counted-major-route fit matches the collision numerator and mileage to the same road scope.
Weakly supported bands and unseen holdout bands are flagged. Bankruptcy and insolvency are inherited area context; their source reporting units are counted explicitly. Shared source areas and historical Cumbria controls stay together in uncertainty blocks.
Retention requires at least 1% mean development deviance improvement, positive temporal and pooled geographic gains, improvement in at least seven geographic holdouts, primary q ≤ 0.05, positive conditional gain intervals and positive exposure sensitivities. It remains a univariate research decision. Joint value, calibration and formal promotion gates are still open.
| Feature | Decision | National banded LSOAs | 2023 gain | Spatial gain | Positive regions | Primary q |
|---|---|---|---|---|---|---|
| Socio-demographic · novel candidate | uncertain | 35,672 | 2.95% | -2.41% | 6/10 | <0.0001 |
| Socio-demographic · novel candidate | uncertain | 35,672 | 1.07% | -2.50% | 6/10 | <0.0001 |
| Socio-demographic · novel candidate | retained for joint testing | 35,672 | 1.07% | 1.94% | 8/10 | <0.0001 |
| Socio-demographic · novel candidate | retained for joint testing | 35,672 | 3.10% | 2.10% | 9/10 | <0.0001 |
| Socio-demographic · novel candidate | retained for joint testing | 35,672 | 4.10% | 4.61% | 9/10 | <0.0001 |
| Socio-demographic · novel candidate | retained for joint testing | 35,672 | 3.61% | 10.28% | 10/10 | <0.0001 |
| Socio-demographic · novel candidate | retained for joint testing | 35,672 | 3.03% | 2.66% | 9/10 | <0.0001 |
| Housing and vehicles · novel candidate | uncertain | 35,672 | 0.04% | -0.53% | 2/10 | 0.0548 |
| Housing and vehicles · novel candidate | retained for joint testing | 35,672 | 3.71% | 8.73% | 10/10 | <0.0001 |
| Housing and vehicles · novel candidate | retained for joint testing | 35,672 | 7.05% | 7.49% | 8/10 | <0.0001 |
| Housing and vehicles · novel candidate | retained for joint testing | 35,672 | 6.66% | 7.21% | 8/10 | <0.0001 |
| Baseline controls · baseline-form diagnostic | uncertain | 35,672 | 0.14% | -0.17% | 5/10 | 0.0005 |
| Baseline controls · baseline-form diagnostic | uncertain | 35,672 | 0.84% | 0.58% | 9/10 | <0.0001 |
| Baseline controls · baseline-form diagnostic | uncertain | 35,672 | 0.62% | 1.18% | 10/10 | <0.0001 |
| Baseline controls · baseline-form diagnostic | uncertain | 35,672 | 0.48% | 0.14% | 5/10 | <0.0001 |
| Road layout · baseline-form diagnostic | uncertain | 35,672 | 1.39% | 0.67% | 8/10 | <0.0001 |
| Road layout · baseline-form diagnostic | uncertain | 35,672 | 0.12% | -0.02% | 7/10 | <0.0001 |
| Road layout · baseline-form diagnostic | uncertain | 35,672 | 0.30% | 0.19% | 7/10 | <0.0001 |
| Road layout · baseline-form diagnostic | uncertain | 35,672 | 0.16% | 0.35% | 7/10 | <0.0001 |
| Road layout · baseline-form diagnostic | uncertain | 35,672 | 0.07% | 0.06% | 6/10 | 0.0015 |
| Road layout · baseline-form diagnostic | uncertain | 35,672 | 0.37% | -0.50% | 7/10 | <0.0001 |
| Weather · novel candidate | uncertain | 35,365 | 1.54% | 1.59% | 4/10 | <0.0001 |
| Socio-demographic · not tested | Not testedDeferred source | 0 | — | — | — | Not tested |
| Socio-demographic · not tested | Not testedDeferred source | 0 | — | — | — | Not tested |
| Socio-demographic · not tested | Not testedDeferred source | 0 | — | — | — | Not tested |
| Socio-demographic · not tested | Not testedDeferred source | 0 | — | — | — | Not tested |
| Road layout · not tested | Not testedDeferred source | 0 | — | — | — | Not tested |
| Road layout · not tested | Not testedDeferred source | 0 | — | — | — | Not tested |
| Traffic and activity · not tested | Not testedDeferred source | 0 | — | — | — | Not tested |
| Traffic and activity · not tested | Not testedDeferred source | 0 | — | — | — | Not tested |
| Traffic and activity · not tested | Not testedDeferred source | 0 | — | — | — | Not tested |
| Traffic and activity · not tested | Not testedDeferred source | 0 | — | — | — | Not tested |
| Traffic and activity · not tested | Not testedDeferred source | 0 | — | — | — | Not tested |
| Traffic and activity · not tested | Not testedDeferred source | 0 | — | — | — | Not tested |
| Traffic and activity · not tested | Not testedDeferred source | 0 | — | — | — | Not tested |
| Weather · not tested | Not testedDeferred source | 0 | — | — | — | Not tested |
| Weather · not tested | Not testedDeferred source | 0 | — | — | — | Not tested |
| Weather · not tested | Not testedDeferred source | 0 | — | — | — | Not tested |
| Housing and vehicles · novel candidate | retained for joint testing | 35,672 | 6.28% | 7.26% | 8/10 | <0.0001 |
| Housing and vehicles · novel candidate | retained for joint testing | 35,672 | 4.29% | 4.56% | 8/10 | <0.0001 |
| Commuting baseline · novel candidate | retained for joint testing | 35,672 | 5.44% | 13.64% | 10/10 | <0.0001 |
| Commuting baseline · novel candidate | uncertain | 35,672 | 0.70% | 1.39% | 5/10 | <0.0001 |
| Commuting baseline · novel candidate | uncertain | 35,672 | 2.10% | -4.81% | 6/10 | <0.0001 |
| Commuting baseline · novel candidate | uncertain | 35,672 | 2.47% | -0.70% | 9/10 | <0.0001 |
| Commuting baseline · novel candidate | uncertain | 35,672 | 0.32% | 0.73% | 7/10 | 0.0435 |
| Transport accessibility · novel candidate | retained for joint testing | 35,672 | 5.64% | 12.99% | 9/10 | <0.0001 |
| Transport accessibility · novel candidate | retained for joint testing | 35,672 | 6.96% | 8.66% | 10/10 | <0.0001 |
| Transport accessibility · novel candidate | retained for joint testing | 35,672 | 4.45% | 7.87% | 8/10 | <0.0001 |
| Transport accessibility · novel candidate | retained for joint testing | 35,672 | 2.63% | 4.02% | 8/10 | <0.0001 |
| Weather · not tested | Not testedDeferred source | 0 | — | — | — | Not tested |
| Weather · not tested | Not testedDeferred source | 0 | — | — | — | Not tested |
| Air quality · novel candidate | uncertain | 35,672 | 0.99% | 0.44% | 5/10 | <0.0001 |
| Air quality · novel candidate | uncertain | 35,672 | 0.54% | -3.43% | 2/10 | 0.0065 |
| Crime · not tested | Not testedBands withheld | 0 | — | — | — | Not tested |
| Crime · not tested | Not testedBands withheld | 0 | — | — | — | Not tested |
| Crime · not tested | Not testedBands withheld | 0 | — | — | — | Not tested |
| Crime · not tested | Not testedBands withheld | 0 | — | — | — | Not tested |
| Speed limits · not tested | Not testedBands withheld | 0 | — | — | — | Not tested |
| Speed limits · not tested | Not testedBands withheld | 0 | — | — | — | Not tested |
| Financial context · novel candidate | uncertain | 35,672 | 0.37% | -1.38% | 3/10 | 0.3516 |
| Financial context · novel candidate | uncertain | 35,672 | 0.51% | -3.18% | 2/10 | 0.1090 |
| Weather · novel candidate | uncertain | 35,365 | 1.06% | 1.85% | 3/10 | <0.0001 |
100 × (Highest level of qualification: No qualifications) / (Highest level of qualification: Total: All usual residents aged 16 years and over)
Source period: 2021-03-21 · Unit: %
Research shortlist gates not met: meanGainAtLeast1Percent, positiveSpatial, sevenGeographies, positiveConditionalIntervals.
Residents aged 16+. Not age-adjusted; age composition is a separate control.
135,903 eligible annual observations; 66,216 in development. 0 otherwise eligible rows lack this feature. Every national LSOA remains in the frozen feature lookup.
| Representation | 2023 next-year gain | Held-out region gain | 2024 diagnostic gain | 2025 exploratory gain |
|---|---|---|---|---|
| Frozen categorical bands | 2.95% | -2.41% | 3.08% | 3.09% |
| Continuous linear | 2.92% | -2.63% | 3.10% | 3.18% |
| Restricted cubic spline | 2.97% | -2.77% | 3.14% | 3.26% |
Best average development representation: categorical. Later-year results do not change the decision. All source editions are retrospective; these are not forecasts using only information available in 2022.
| Band | National LSOAs | 2023 collisions / miles | 2023 frequency / 100k | Adjusted RR (95% interval) | Effect sample support |
|---|---|---|---|---|---|
| D1 | 3,569 | 10,338 / 23,876,791,713 | 0.0433 | 0.824 (0.769–0.884) | 3,228 LSOAs; 20,842 collisions; 140 blocks |
| D2 | 3,567 | 10,136 / 38,386,212,464 | 0.0264 | 0.876 (0.832–0.923) | 3,315 LSOAs; 20,666 collisions; 155 blocks |
| D3 | 3,566 | 9,414 / 37,718,589,632 | 0.0250 | 0.884 (0.840–0.929) | 3,282 LSOAs; 18,993 collisions; 159 blocks |
| D4 | 3,567 | 10,014 / 36,812,137,280 | 0.0272 | 0.954 (0.913–0.998) | 3,332 LSOAs; 20,046 collisions; 161 blocks |
| D5 | 3,567 | 9,432 / 32,317,929,740 | 0.0292 | 1.000 (1.000–1.000) | 3,321 LSOAs; 18,975 collisions; 162 blocks |
| D6 | 3,568 | 8,899 / 27,809,035,505 | 0.0320 | 1.066 (1.024–1.110) | 3,329 LSOAs; 18,241 collisions; 162 blocks |
| D7 | 3,567 | 8,614 / 23,201,143,207 | 0.0371 | 1.100 (1.052–1.150) | 3,327 LSOAs; 17,560 collisions; 162 blocks |
| D8 | 3,567 | 8,457 / 22,262,814,985 | 0.0380 | 1.115 (1.059–1.173) | 3,347 LSOAs; 17,225 collisions; 160 blocks |
| D9 | 3,567 | 8,268 / 17,232,277,597 | 0.0480 | 1.274 (1.206–1.347) | 3,339 LSOAs; 16,608 collisions; 153 blocks |
| D10 | 3,567 | 9,131 / 16,592,667,877 | 0.0550 | 1.586 (1.480–1.699) | 3,288 LSOAs; 18,458 collisions; 136 blocks |
| 2023 next-year check | Deviance gain | Conditional 95% interval | O/E | Rows |
|---|---|---|---|---|
| Fitted-flow mileage | 2.95% | 2.17% to 3.80% | 0.951 | 33,108 |
| Traffic-group mileage | 3.02% | 2.20% to 3.92% | 0.953 | 33,108 |
| Road-length mileage | 2.90% | 2.10% to 3.76% | 0.955 | 33,108 |
| Counted major routes | 0.05% | -0.98% to 1.04% | 0.967 | 5,098 |
Intervals resample 1,000 fixed-prediction spatial blocks. They omit refitting, selection and exposure uncertainty, so they do not satisfy the E8 promotion gate. Subgroup effects are exploratory; only the primary categorical test enters its declared FDR family.
All fitted model parameters · Paired validation block totals · Original feature source
The next phase can test retained and justified uncertain features jointly, assess overlap within feature families and apply shrinkage. Individual significance is insufficient: correlated variables may add little once combined. Earlier-collision history remains a demanding additional comparator.
No composite score or pricing factor is released here. Formal promotion still requires uncertainty that refits models and selection, improved calibration, credible exposure and untouched outcome or insurer-claims evidence. Census, air quality, insolvency, climate normals and April 2026 roads have different vintages; this release does not certify historical underwriting availability.
Evidence cards include source definitions, coverage, raw and adjusted bands, all annual and regional effects, prediction metrics and decision gates. The reproduction kit contains exact frozen inputs and code; no new national quantiles are calculated.
| Download | Purpose | Size |
|---|---|---|
| All 62 candidate decisions | Coverage, test family, primary q, development gains and research disposition. | 0.03 MB |
| All predictive comparisons | Linear, spline and categorical development tests; exploratory later years and mileage sensitivities. | 0.37 MB |
| Adjusted band effects | Pooled, annual, region and urban/rural fits; cluster intervals and support flags. | 0.29 MB |
| Raw national-band frequencies | National membership, eligible collision counts, summed miles and frequency per 100,000 miles by year/scenario. | 0.09 MB |
| Held-out subgroup diagnostics | Year, region, urban/rural, better-observed mileage and band calibration. | 0.55 MB |
| Continuous effect curves | Linear and restricted-cubic-spline incremental effects with conditional cluster intervals. | 0.11 MB |
| Independent uncertainty blocks | Every LSOA, authority and source reporting area; shared-source and Cumbria unions. | 0.08 MB |
| Complete reproduction kit | Pinned inputs and executable code to refit all models and rebuild all numerical evidence cards. | 35.17 MB |
Frozen screening rulesMethods & reproductionPinned inputsAll output checksumsVerification record
Browser interaction QA was unavailable in this build environment. Statistical engine checks, numerical reproduction, server-rendered page checks, the production build and published download checks are recorded separately.