# PRE-REGISTRATION — Housing Delivery Test 2025 × Major-Dwellings Approval Behaviour (England) **Status: FROZEN at the time stamp below. Written before any PS2 decision value was read.** Frozen: Wednesday 2 September 2026, ~18:50 BST. Author: Claude, for Mark Broome / PlanningLens. Any departure from this document during analysis is to be recorded in §12 ("Amendments") with a reason, not silently applied. --- ## 1. Question and hypotheses **Question.** Does the rate at which English local planning authorities grant major residential planning applications relate to whether they meet the Housing Delivery Test? **H0 (null):** authority-level major-dwellings approval rate is not associated with HDT 2025 performance. **H1a:** authorities with lower HDT results approve a lower share of major-dwellings applications (refusal tracks under-delivery). **H1b:** authorities with lower HDT results approve a *higher* share (the "approving but not delivering" pattern). **Policy hypothesis to be tested, not assumed:** if authorities failing the HDT grant major-dwellings applications at essentially the same rate as those passing it, one of the test's punitive mechanisms (presumption in favour of sustainable development below 75%; 20% buffer below 85%; action plan below 95%) targets a planning behaviour that does not distinguish successful from unsuccessful delivery areas. This is an association claim only. **Not claimed under any result:** causation in either direction; "councils approved the homes but developers failed to build them"; anything about committee versus delegated decisions; that the official percentage is the approval rate for brand-new schemes (see §4). --- ## 2. Data sources (fixed) | Item | Source | File / field | |---|---|---| | Housing Delivery Test 2025 | MHCLG, GOV.UK, published 17 Aug 2026 | `HDT_results_2025.ods` → `hdt_2025_clean.csv`; fields `ons_code`, `lpa`, `hdt_pct` (measurement × 100), `consequence`, `req_total`, `del_total`. 301 numeric results. | | Residential planning decisions | MHCLG PS2 open data, "District Planning Application Statistics (PS2), England, 1 April 1979 – 31 March 2026", file `PS2_data_-_open_data_table__202603_.csv` (server last-modified 18 Jun 2026, 61.3 MB; header on row 3) | Key columns `LPACD` (ONS code), `LPANM`, `Region`, `Quarter` (calendar label, e.g. "2022 Q2"). Measures: `Total decisions; major dwellings (all)`, `Total granted; major dwellings (all)`, `Total refused; major dwellings (all)`; secondary `…; minor dwellings (all)`. | | Green Belt (sensitivity covariate) | MHCLG "Local authority green belt statistics for England", latest annual LPA-level area table (2024/25 if available) | Green Belt area (hectares) per LPA. To be fetched after freezing; source and vintage recorded in §12. | | PlanningLens scheme layer | `hdt_feasibility_20260902.py` v2026-09-02h output (run 3, 2 Sep 18:30) | `lpa_metrics.csv`: per-LPA major decisions, approval %, named-scheme access via council databases. | **Join key:** `LPACD` = HDT `ons_code` (E06/E07/E08/E09; E51 development corporations if present in both). No name matching. --- ## 3. Window (fixed) Decisions in PS2 quarters **2022 Q2, 2022 Q3, 2022 Q4, 2023 Q1, 2023 Q2, 2023 Q3, 2023 Q4, 2024 Q1, 2024 Q2, 2024 Q3, 2024 Q4, 2025 Q1** — i.e. 1 April 2022 to 31 March 2025, the exact HDT 2025 measurement window (FY2022/23–2024/25). Twelve quarters, summed per authority. --- ## 4. Cohort and outcome — exactly as MHCLG reports it **Cohort:** "major dwellings (all)" as defined in the PS2 data dictionary: development involving the provision of 10 or more dwellings, or a site of ≥0.5 ha where the number is not known. "(all)" = including applications subject to performance agreements, extensions of time and EIA. The dwellings category is assigned by principal use. **Approval rate:** `granted / decisions` per authority over the twelve quarters, matching MHCLG's published "% granted". Refusal rate = `refused / decisions`. Consistency check: `decisions − (granted + refused)` recorded per authority; any authority where the discrepancy exceeds 2% of decisions is flagged in output and retained (MHCLG's own totals are used). **Documented limitation — reserved matters and s73.** PS2 counts every decision on an application of the type, including approvals of reserved matters and section 73 (variation of condition) applications, which are predominantly granted. The official "% granted" for major dwellings is therefore *not* the approval rate for brand-new schemes; it is systematically higher, and the inflation is assumed uniform across authorities for the cross-sectional comparison. The PlanningLens scheme-level layer, which excludes RM and s73 by construction, is used to illustrate the distinction for named authorities; it is not used for the national statistic (its recall is ~25–50%, see the feasibility report). **Missing returns:** an authority is included only if it has a non-blank `Total decisions; major dwellings (all)` value in at least **10 of the 12** quarters. The number of quarters present is recorded per authority and the count of excluded authorities reported. --- ## 5. Sample-size floor (fixed) - **Primary:** ≥ **20** major-dwellings decisions over the window. - **Sensitivity:** ≥ 40. - Minor dwellings (secondary analysis): ≥ 60. Authorities below the floor are listed, not analysed. --- ## 6. HDT bands (fixed) **Primary (MHCLG consequence bands):** <75% (presumption) · 75–<85% (20% buffer) · 85–<95% (action plan) · ≥95% (no consequence). **Secondary (analysis bands):** <75% · 75–<95% · ≥95%; the ≥100% subset also shown. HDT % is used as a continuous variable for correlations; bands are for description only. --- ## 7. Statistics (fixed) 1. **Primary measure:** Spearman rank correlation ρ between HDT % and major-dwellings approval % across eligible authorities. 2. **Uncertainty:** 95% confidence interval for ρ by bootstrap (2,000 resamples of authorities, percentile method); Fisher-z interval also reported. 3. **Interpretation rules, fixed in advance:** - |ρ| < 0.15 is the pre-defined region of **no meaningful relationship**. - The phrase "no meaningful relationship" may be used only if the 95% CI lies entirely within (−0.30, +0.30) **and** the point estimate is within ±0.15. If the CI also lies within (−0.15, +0.15) this is reported as a strong null. - 0.15 ≤ |ρ| < 0.30 = weak; ≥ 0.30 = moderate or stronger. Direction reported. - A non-significant result is **not** described as proof of no relationship; the CI is always quoted alongside the point estimate. 4. **Band description:** median approval % per HDT band with bootstrap 95% CI (2,000 resamples) and inter-quartile range; number of authorities per band. 5. **Band contrast:** difference in median approval % between the <75% band and the ≥95% band, with bootstrap 95% CI. 6. Decisions-weighted approval % per band (pooled granted / pooled decisions) also reported, so the reader sees both the authority-level and the application-level view. 7. No thresholds on approval level (e.g. 85% or 90%) are treated as pass/fail criteria for the finding. --- ## 8. Analysis cuts (fixed, all reported) 1. **England, all eligible authorities.** 2. **England excluding London** (Region = London / ONS code E09) — the headline cut for national description. 3. **London only** — reported as a separate panel. 4. **Green Belt sensitivity:** (a) authorities with any designated Green Belt vs none; (b) if LPA land area is obtainable, authorities with Green Belt ≥ 25% of land area vs < 25%. Correlations and band medians within each group. 5. **Extreme-HDT sensitivity:** excluding authorities with HDT > 200%. 6. **Authority-type sensitivity:** excluding development corporations and any non-district authority present in both datasets. 7. **Sample-floor sensitivity:** repeat (1)–(3) at ≥ 40 decisions. **Secondary outcome:** the same cuts for **minor dwellings (all)** with the ≥ 60 floor. --- ## 9. Anomaly identification (rule-based, all listed) Within eligible authorities (primary cut, England excluding London; London listed separately): - **A. Low HDT, high approval:** HDT < 75% and major approval % in the top quartile of eligible authorities. - **B. High HDT, low approval:** HDT ≥ 100% and major approval % in the bottom quartile. - **C. Largest requirements:** the 20 authorities with the largest `req_total`, with their approval % and band. All qualifying authorities are listed; none are dropped for narrative convenience. --- ## 10. PlanningLens scheme layer — eligibility rule Named schemes and stated unit counts are drawn from PlanningLens **only** for authorities where, over the same window, the PlanningLens major-new cohort (v02h) counts at least **40%** of the PS2 major-dwellings decisions for that authority, and which are not among the 18 authorities whose descriptions PlanningLens cannot currently read. The PlanningLens/PS2 ratio is disclosed next to every named authority. Authorities failing this rule are described from PS2 only. --- ## 11. Outputs - `hdt_ps2_lpa.csv` — one row per matched authority: codes, HDT %, band, consequence, quarters present, major decisions/granted/refused/approval %, minor equivalents, London flag, Green Belt flag(s), eligibility flags, PlanningLens major count and ratio. - `hdt_ps2_summary.md` — all statistics in §7 for every cut in §8, anomaly lists in §9, exclusion counts, consistency-check counts. - No headline, narrative or press language is written until Mark has read the numbers. --- ## 12. Amendments (append-only; empty at freeze) Frozen text hash at freeze (before this section was appended): sha256 2ce6a5780b73f32f403a704e9504d592e8f5320d086b80fc16606033c53f1608. 1. 2 Sep 2026 ~19:05 — PS2 CSV is Windows-1252 encoded, not UTF-8; read with `encoding="cp1252"`. Non-substantive (file parsing only). 2. 2 Sep 2026 ~19:05 — Green Belt source fixed as MHCLG "Local authority Green Belt statistics for England: 2024 to 2025" accompanying tables (`Live_Tables_-_Green_Belt_Statistics_2024-25.ods`, sheet `Area_by_LA`: authority code, Green Belt hectares, total land area at 31 Dec 2024, percentage). Authorities absent from the sheet treated as 0 ha. Both pre-registered Green Belt cuts (any/none; ≥25% of land) were therefore run as specified. 3. 2 Sep 2026 ~19:05 — Additional reporting, not a change: the Green Belt any/none cut was also reported for England excluding London (cut 4a′) alongside the pre-registered all-England version. No pre-registered cut was dropped or altered. 4. 3 Sep 2026 — Identifier correction, made after the frozen results had been read. MHCLG's HDT 2025 and PS2 files code Sheffield as E08000039 and Barnsley as E08000038; MHCLG's Green Belt live table (amendment 2) codes the same two authorities as E08000019 and E08000016. The frozen run of 2 Sep therefore joined neither Green Belt row and, under amendment 2's absent-authority rule, treated both as 0 ha. The analysis script now maps the two Green Belt codes to the HDT/PS2 codes before the join (`GB_CODE_CROSSWALK` in hdt_ps2_analysis_20260903.py; no other line differs from the 2 Sep script, which reproduced its frozen outputs exactly before the change) and both scripts were re-run and re-frozen on 3 Sep 2026 as HDT_PS2_RESULTS_20260903.md; the 2 Sep results remain published alongside for comparison. Effect: Sheffield (24.6%) stays in the <25% group; Barnsley (68%, HDT 75%) moves to the ≥25% group, which becomes 85 authorities (was 84) against 144 (was 145). Barnsley is in neither the <75% nor the ≥95% band, so all four headline medians and their sample sizes are unchanged (70.8 and 85.0, n 27 and 46; 83.7 and 83.2, n 35 and 79), as is the <25% decisions-weighted comparison (81.0 vs 81.0). Spearman rho: ≥25% group +0.449 → +0.442; <25% group +0.130 → +0.135; exploratory permutation p (floor 20) 0.0112 → 0.0158; rank-regression interaction +0.364 → +0.338. Bootstrap intervals downstream of the changed group sizes shift with the RNG stream and are reported as re-frozen. The pre-registered any/none Green Belt cut (4a) also changes composition (150/100, was 148/102). No pre-registered cut, threshold, rule or interpretation criterion was altered.