{
  "id": "20mph-phase1-v1",
  "frozenAt": "2026-11-03",
  "status": "Versioned study specification; feasibility release",
  "priorKnowledge": "Written after inspecting published outcomes and existing site studies. Not a prospective preregistration or unseen holdout claim. Later amendments require a new version and reason.",
  "question": "What changes in police-recorded injury harm, speeds and journey times follow a 30-to-20mph limit change, compared with plausible outcomes without that change?",
  "geography": "Wales first; English comparison areas and roads subject to treatment-history checks",
  "policyDate": "2023-09-17",
  "primaryPre": [
    "2015-01-01/2019-12-31",
    "2022-01-01/2023-08-31"
  ],
  "transitionExcluded": "2023-09",
  "primaryPost": "2023-10-01/2025-12-31",
  "postFullMonths": 27,
  "pandemic": "2020–2021 excluded from main analysis; separate sensitivity",
  "provisional": "Welsh quarterly series through Q2 2026 kept separate from final microdata; no blending",
  "primaryOutcomes": [
    "Police-reported injury collision counts on a fixed road network",
    "Total casualties on the same network"
  ],
  "secondaryOutcomes": [
    "Fatalities separately; adjusted serious injuries only when force/year adjustments are verified",
    "Collisions per 100,000 vehicle miles where exposure matches roads and period",
    "Measured speeds at retained TfW sites",
    "Journey durations and 5th–95th percentile spread on the same routes, direction and seasonal peak"
  ],
  "treatments": [
    "30-to-20",
    "unchanged-30",
    "already-20",
    "traffic-calmed-20",
    "part-time-20",
    "reverted-to-30",
    "early-pilot"
  ],
  "roadPanel": "Independent complete road inventory including zero-collision sections, split at legal limit boundaries; exact dates and versioned geometry required. Retain uncertain matches and unmatched collisions. Collision speed labels never define treatment or full network.",
  "areaFallback": "All-road counts as the primary descriptive feasibility path. Recorded20/30 combined remains a category proxy with an explicit reconciliation warning. Verify stable boundaries and historical coverage before a controlled panel.",
  "identification": {
    "road": "Match using pre-policy road class, urban setting, junctions, baseline speed/traffic, casualty history and deprivation. Difference-in-differences/event study; use cohort-aware methods for different adoption dates.",
    "area": "Weighted English comparison chosen on pre-policy data only. Screen local20mph adoption, reporting changes and spillovers before freezing donors. Welsh30mph exceptions are sensitivity controls only.",
    "inference": "One national policy shock: do not treat road sections as independent policy experiments. Assess time/spatial dependence with block/placebo inference; no causal claim from a before/after chart.",
    "exposure": "Present total harm alongside matched mileage rates; traffic may mediate the policy effect. AADF supports annual estimates, not observed monthly road traffic."
  },
  "recordingBreaks": [
    {
      "force": "Dyfed-Powys",
      "date": "2023-05",
      "event": "Injury-based severity recording"
    },
    {
      "force": "South Wales",
      "date": "2025-11",
      "event": "Injury-based severity recording"
    }
  ],
  "sensitivity": [
    "Pre-trends and placebo dates/areas",
    "Alternative comparison pools and leave-area-out checks",
    "Pandemic windows and secular trends",
    "Traffic diversion: all roads and adjacent roads",
    "Police recording changes and concurrent road schemes",
    "Unknown legal histories, spatial matches and exposure quality",
    "Poisson/overdispersion and block uncertainty"
  ],
  "subgroups": {
    "roadUsers": [
      "pedestrian",
      "cyclist",
      "motorcyclist",
      "car occupant",
      "other"
    ],
    "ages": [
      "0–15",
      "16–24",
      "25–64",
      "65+"
    ],
    "rules": "Pre-specify broad groups. Estimate support and detectable effects before subgroup claims; partial pooling for sparse groups. Reversals exploratory until sufficient follow-up."
  },
  "releaseGates": {
    "phase2": "Historical imports, stable geography and reconciliation displayed; descriptive labels and archived inputs required.",
    "phase3": "Exact dated road inventory OR a verified area counterfactual; pretrend/placebo diagnostics; dependence-aware power/inference; exposure for any rate claim.",
    "phase4": "Adequate subgroup support; separate approval.",
    "phase5": "Cost-benefit only after credible effect and time/exposure evidence; no insurance claims or premium conclusion without insurer data."
  },
  "dataDictionary": {
    "collision": "One injury collision; collision_index unique",
    "vehicle": "One vehicle per collision_index + vehicle_reference",
    "casualty": "One casualty per collision_index + vehicle_reference + casualty_reference",
    "country": "ONS district at collision location; alternative force/highway assignments audited",
    "speedLimit": "Recorded incident field; not a legal treatment history",
    "missing": "Unknown stays unknown; missing exposure does not become zero",
    "zero": "Zero events only valid after complete inventory/time coverage confirmed"
  },
  "nextDecision": "Approve Phase2 descriptive pilot separately after reviewing phase1 findings. Causal modelling and costs remain later phases."
}
