Given What You Received

About this analysis. Abre Etteh is an ARB-registered architect, a member of a London borough design review panel, and the author of the London Borough of Merton's adopted small sites design guide — one of three such documents adopted in London. Perfect Scale serves the small-sites ecosystem: the same dataset and the same method underpin work for applicants appraising sites and for authorities examining their own decision record and policies. The analysis is published under a single auditable method; Appendix A states the conditions under which its reading would be falsified, and any London borough may request its own case-level extract (§8.5).

Executive summary

London does not operate a single planning system for small housing developments; it operates more than thirty. Across twelve thousand decisions on schemes of one to nine homes, the gap between the most permissive and most restrictive boroughs runs to roughly forty-three percentage points, once you account for the kinds of applications each borough actually receives. A small housing scheme that would typically be approved in Richmond stands a good chance of refusal in Havering, not because it is a worse scheme, but because the two boroughs produce systematically different results on comparable applications.

The comparison is adjusted for caseload mix. Boroughs differ in the types of sites they receive (conversions versus new-builds, conservation area sites versus open-plan suburban plots, schemes of two units versus seven) and a borough that receives genuinely more difficult applications should, all else equal, refuse more of them. The differential accounts for this by asking what London-average decision behaviour would have produced on each borough's own caseload, then measuring how far the actual outcome fell above or below that expectation. It is a measure of outcome divergence under comparable conditions, not a direct measure of a Local Authority's intent. It sweeps up unobserved differences in preparation quality, consultant sophistication and pre-application filtering alongside any genuine difference in how boroughs decide. That ambiguity is addressed throughout the essay.

A small-site scheme that London-average decision behaviour predicts would be approved a little under sixty per cent of the time is consented at around seventy-nine per cent in Richmond and a little under thirty per cent in Havering. The clearest pattern is a positive correlation between a borough's outcome differential and its house prices: the more affordable the borough, the more its outcomes lag the rate its caseload mix predicts. This describes the outcome of the planning regime, and on this evidence does not yet establish its intent. Two readings of the correlation are both consistent with the data: a decision-culture reading, in which more affordable boroughs decide harder than their caseload predicts, and an applicant-quality reading, in which they receive less-prepared applications because preparation pays less at lower values. The essay argues the truth is a weighted mix the dataset cannot yet cleanly separate (§5). The word "regressive" is used throughout in its technical economic sense — the distributional shape of the outcome — rather than as a moral verdict on any borough or official.

The convergence trend visible in 2023–2025 constrains how the cross-section can be read. Over the visible window the wealth–approval correlation has weakened from +0.66 in 2023 to +0.55 in 2025, and the spread between the four hardest and five most permissive boroughs has narrowed from roughly 55 percentage points to roughly 43. London-wide approval rose from 52% in 2023 to 61% in 2025, with a partial 2026 reversion to 57%. This trend indicates that the system moves under political pressure and partly reverts. Our reading is that the December 2024 National Planning Policy Framework (NPPF) revision pressed hardest on underdelivering councils, which fall disproportionately among the affordable outer boroughs. The convergence and the additional homes are the result. The underlying regulatory cost asymmetry has not changed, and the 2026 reversion is consistent with the lift being a one-cycle intervention rather than a durable shift.

What the evidence supports and what it does not

Supports. A price-skewed cross-section, in which outcome differentials track house prices and the extremes are robustly identified. A wealth–approval correlation that excludes zero. A statistically large, Croydon-specific shift across a political cycle that peers did not match. A withdrawal-rate asymmetry consistent with expensive boroughs filtering invisibly while affordable boroughs filter visibly. A rising share of insufficient-information refusals in affordable boroughs through 2023–2025. A convergence trend in the same window consistent with central political pressure mattering and with the regime not being fixed. A within-agent cross-borough regression showing borough rankings stable (Spearman ρ = 1.0) with and without agent fixed effects — directional support for the decision-culture reading, though it does not close the debate (detail in A.9).

Does not support. Identification of which competing reading of the cross-section dominates. A causal claim that the revocation of Supplementary Planning Document 2 (SPD2), Croydon's small-sites design guide, alone produced the post-revocation Croydon level. A claim that the regressive cost asymmetry has been quantified at borough level. An attribution of intent to any borough or official. A forecast that the convergence trend will reverse rather than persist.

The policy frame we draw from this evidence is a transparency regime rather than a central-override regime: publish the differential annually, give each borough the means and the forum to account for a large differential in the terms of its adopted local plan, weigh compliance-cost decisions against their distributional consequences, and let democratic process do the corrective work where the differential reflects a legitimate local choice. The s.24(7) conformity opinion remains the principal lever where a borough's local plan has materially diverged from the strategic policy of the London Plan, and not a routine override of borough discretion where no such divergence exists.

What this yields, by reader. For a local authority, it offers a mirror — a measured account of what its chosen posture is producing, framed as a question it can answer on its own terms, with the evidence to answer it (§8). For central government, it marks where transparency and a single proportionality test could remove distributionally-skewed cost without overriding a legitimate local choice (§8). For a developer, it changes two decisions: which boroughs to weight in acquisition, and where to spend preparation effort and scrutiny once a site is in play (§9). None of it tells any borough to become more permissive; all of it asks each to be able to say what its chosen level of permissiveness, and the burden behind it, is actually achieving.

§1. The finding

Comparing raw approval rates between boroughs is close to useless, and worse than useless if it informs policy. A borough that approves only 40% of applications may be strict, or it may simply receive a harder set of applications. Raw rates conflate the difficulty of the decision with the difficulty of the caseload. They punish boroughs for the applications that happen to land on their desks.

The comparison that does the analytical work removes the caseload from the picture. For every application in the dataset we ask what its probability of approval would have been under London-average decision behaviour, given its site type, transport accessibility (PTAL band), conservation-area status, and the number of units proposed. Summed across a borough's caseload, this yields a mix-expected approval rate. The outcome differential — more fully, the residual outcome differential after mix controls — is the borough's actual approval rate minus this expectation.

Reading the outcome differential

A differential of zero means the borough's outcomes match the rate the London average would have produced on its caseload. A negative differential means the borough produces lower approval rates than its caseload mix predicts — consistent with the borough deciding harder, but also consistent with the borough receiving systematically lower-quality applications on the unobserved dimensions the model cannot capture. A positive differential means the borough produces higher approval rates than its caseload mix predicts — consistent with the borough deciding more permissively or with the borough receiving systematically higher-quality applications.

The differential gives each borough full credit for receiving difficult sites along the four observable dimensions. It does not give them credit (or take it away) for unobserved differences in scheme quality, applicant sophistication, or pre-application filtering. The differential is therefore a descriptive map of outcome divergence under like-for-like observable mix; it is not a measurement of how the borough decides.

§2. The ranking

The hardest and most permissive boroughs in the table both deserve scrutiny. Below are the extremes.

The five hardest boroughs

The five most permissive boroughs

Two patterns leap off the page. First, the hardest boroughs (Havering, Croydon, Barking & Dagenham, Waltham Forest, Newham) are mostly outer and east London — the more affordable, eastern and southern names dominate, with Waltham Forest the one mid-priced exception. Second, the most permissive boroughs (Richmond, Bexley, Kensington & Chelsea, Hammersmith & Fulham, Ealing) pair the expected wealthy inner-London names with two more affordable outer boroughs, Bexley and Ealing. The outcome differentials track money on average, with informative exceptions at both ends — Bexley above all.

§3. The delivery matrix

The differential answers one question: how does a borough's small-sites approval rate compare with what its caseload mix predicts? It is silent on a second that matters at least as much: to what end? A borough can sit well below expectation on small sites and still deliver homes in volume through other parts of its pipeline; another can sit below expectation with little to show for it. The ranking measures decision outcomes rather than housing output, so it cannot separate the two. That separation needs a second axis: how much housing a borough delivers. The differential remains the primary measure here, built on twelve thousand of the borough's own decisions; the delivery axis is a complicating layer laid beside it, and the pairing is descriptive rather than causal.

The measure we wanted and the measure we can get

The coherent partner to a small-sites permissiveness axis would be small-sites completions per borough. That number is not published. The Greater London Authority (GLA) reports completions by borough (all sites) and completions by site size (London-wide), but never crosses the two; the London Development Database, which underpins both, is known to undercount small-site completions because small schemes are reported inconsistently. The gap is itself telling: the system can measure small-sites permission to a decimal place and small-sites delivery hardly at all. In its place we use the Housing Delivery Test (HDT), which is central government's own measure of the homes a borough delivered against the homes it was required to deliver, published per borough and carrying statutory consequences (presumption, buffer, action plan). It is an all-sites measure, and a lagged one, as the published figures span roughly 2019–2022, while the differential spans 2022–2026. So the matrix deliberately pairs a small-sites permissiveness axis with a whole-borough delivery axis measured at a different time. It cannot answer whether a low small-sites differential reduces small-sites delivery; it shows what a borough's small-sites posture sits alongside in its overall delivery record.

The hardest boroughs split on delivery

The boroughs with the largest negative differentials show a correlation with delivery. Havering (−23.0pp differential, 61% of required delivery), Barking & Dagenham (−18.0pp, 66%) and Newham (−15.6pp, 61%) pair large negative differentials with delivery well below the level required of them. Croydon and Waltham Forest sit at comparable levels of small-sites restriction (−21.7pp and −16.3pp) yet clear their delivery test comfortably — Croydon at 160% of required, Waltham Forest at 119% — by delivering through strategic allocations rather than small sites. The same small-sites posture can sit beside opposite delivery records, so the label "hard" covers more than one situation.

The permissive end splits also

The same division repeats at the top of the differential. Hammersmith & Fulham (+16.2pp, 143%) and Bexley (+17.3pp, 106%) combine above-expectation small-sites approval with delivery above requirement. Richmond (+19.9pp, 60%), Kensington & Chelsea (+17.1pp, 63%) and Camden (+10.0pp, 53%) approve small sites more readily than their caseload predicts yet still deliver far below requirement. Permission at the front of the pipeline does not guarantee homes at the end of it: build-out lag, scheme viability, land supply and a borough's small absolute size all sit between a consent and a completion. A permissive planning department is not, on its own, a delivery engine.

Reading the four quadrants. The matrix sorts boroughs into four cells. Restrictive and under-delivering (upper-left; 11 boroughs): the largest negative differentials together with a delivery shortfall, and no sign the restriction corresponds to homes secured elsewhere. Restrictive and productive (upper-right; 6): hard on small sites while delivering through larger ones — the posture a borough can most readily defend. Permissive and under-delivering (lower-left; 10): generous on approval but low on delivery, which points the supply question away from the planning department. Permissive and productive (lower-right; 5). Borough membership is in the appendix table.

How to read this for each borough. The matrix is a framing device. A local authority can ask which quadrant it occupies, and whether its small-sites posture helps explain its delivery record or its delivery is coming from somewhere its small-sites posture has little to do with. A developer can ask whether a borough combines an attractive risk-adjusted permission environment at the front of the pipeline with credible delivery prospects at the end of it and, if not, which of the two is the binding constraint on a scheme like theirs. The same cell can arise from different mechanisms, so the quadrant points each reader to the follow-up question worth their time rather than dictating an answer. The implications sections (§8 for authorities, §9 for developers) take those two questions in turn.

What the matrix can and cannot show. Its delivery axis is all-sites while its permissiveness axis is small-sites, so a "restrictive and productive" borough is delivering through larger sites. The matrix locates that delivery elsewhere rather than vindicating its small-sites posture. The HDT figures span roughly 2019–2022 and the differential spans 2022–2026; delivery lags the decisions that produce it, so the two axes are not contemporaneous and a borough's position can move. Borough-level small-sites completions are not published consistently or crossed centrally by the GLA but individual borough monitoring does report them (Russell Curtis cites Croydon small-site net completions rising from 770 in 2012/13–2016/17 to 1,965 in 2017/18–2021/22, with Barnet next at 710), but there is no clean, comparable London-wide series, and that absence is itself an accountability gap. Quadrant placement is descriptive: delivery is shaped by viability, land supply, build-out rates and borough size as much as by planning posture.

What better data would change. Were borough-level small-sites completions ever published, the delivery axis could be rebuilt on a like-for-like, small-sites basis, and the matrix would move. The restrictive and productive quadrant would likely thin, as boroughs that clear their all-sites test through strategic allocations (Croydon, Waltham Forest) lose the credit that currently lifts them. The restrictive and under-delivering reading of the affordable outer boroughs would, if anything, sharpen. And the permissive and under-delivering quadrant would finally become testable, turning on whether permissive small-sites approval actually converts into small-sites completions. The matrix's qualitative message — that permissiveness and delivery are distinct — survives that data swap; the specific quadrant placements, especially for the all-sites beneficiaries, may not.

Can the delivery axis be made small-sites-specific?

The obvious objection to the matrix is that its delivery axis is all-sites while its permissiveness axis is small-sites. Can the Housing Delivery Test be disaggregated to fix that? On the available data, no. The HDT is a single net-completions total per borough, with no site-size dimension to split; apportioning it by each borough’s small-sites target share is circular, returning the overall HDT ratio for the small-sites slice and adding nothing; and a genuine scheme-level split would need completions records carrying unit counts or site areas — the London Development Database, which undercounts small schemes, or individual borough monitoring, which is not compiled on a comparable London-wide basis. There is currently no direct route from the published HDT to a small-sites completions figure, though we are looking into proxies to approximate this missing data.

What the data does support is a reliance weighting, shown as the SS % of target column in the appendix table. The London Plan sets each borough a small-sites target (sites below 0.25 hectares) and that target is a very different share of the whole from one borough to the next: from about 12% of Barking & Dagenham’s requirement to 84% of Richmond’s, median roughly 30%. The same HDT result therefore means different things in different boroughs. Richmond fails its delivery test while approving small sites readily and depending on them for 84% of its target, so its shortfall is almost entirely a small-sites build-out problem, not a consenting one. Barking & Dagenham, Greenwich and Newham also fail, but lean on small sites for an eighth to a sixth of their targets, meaning that their shortfall sits with larger sites, and their restrictive small-sites posture, genuine as it is, is not the main driver of the underperformance. The reliance share does not measure small-sites delivery; it tells you how much of a borough’s delivery record a small-sites reading can claim.

§4. The affordability problem

Plot each borough's outcome differential against its median ward price and the relationship is positive: the more affordable the borough, the more its outcomes lag the rate its caseload mix predicts. Across the 30 boroughs with reliable price coverage, differential and price correlate at roughly +0.5 (Pearson and Spearman) — a moderate relationship, with the full statistical detail in A.4.

The pattern is descriptively skewed by price. Whether that shape is produced by borough decision behaviour, by applicant-quality variation across price bands, or by some weighted combination of the two, is a separate question taken up in §5. The correlation describes the outcome; on its own it does not reveal the mechanism.

Small sites (backland plots, conversions, modest infill, demolish-and-rebuild) are the part of the housing pipeline most accessible to small and medium builders. SME builders' share of new homes has fallen from roughly 40% in the 1980s to around 10–12% today.¹ The Letwin Review identified small-site delivery as the natural province of less well-capitalised builders.² The London Plan 2021 sets a ten-year small-sites target of 119,250 net dwellings, about 23% of the capital's total housing target.³

So when the planning regime produces its lowest approval outcomes (relative to caseload mix) in exactly the more affordable boroughs, it produces a regressive outcome along an axis that matters for affordability, regardless of the underlying mechanism. This is a descriptive observation, and does not require any borough or official to have intended it.

The Bexley–Havering pair: a counterexample

The clearest internal challenge to a strict price-determines-outcome reading lies in two boroughs with effectively identical house prices.

Three thousand pounds apart on price. Forty percentage points apart on differential.

Bexley and Havering are similar on several observable dimensions beyond price. Both are outer-London boroughs with comparable population density bands, similar PTAL profiles, ward-level deprivation profiles broadly within the same decile band, and Conservative-aligned council compositions in recent cycles. The variables we can observe align them more closely than the 40pp outcome gap suggests. Whatever explains that gap, it is not price, density band, transport accessibility, or basic demographic profile alone.

This does not prove that borough decision-culture is the dominant mechanism: a single-pair comparison, with no control for unobserved scheme quality at either end, is suggestive rather than conclusive. But a £260k outer-London borough that produces +17pp of positive differential, sitting alongside a £263k outer-London borough that produces −23pp, is uncomfortable for any reading in which the observable variables mechanically determine outcomes.

The affordable-and-permissive trio

These three boroughs are an indicative counterexample: the harsh outcome pattern is not inevitable in the affordability band, and a different outcome is observed within the same statutory framework. The mechanism producing the difference between Bexley and Havering is the unresolved question.

§5. What the differential is (and is not) evidence of

Before drawing policy conclusions, the analysis has to confront what this measure cannot do. The outcome differential controls for site type, PTAL, conservation, unit count. It cannot control for unobserved scheme quality, applicant sophistication, agent quality, or pre-application filtering. There are at least four families of unobserved variable that could account for part of the differential.

The agent-quality and repeat-player confound

The most serious unobserved confound is applicant-side. Planning consultants, architects, and pre-application advisers are not randomly distributed across London boroughs. The professional services tier concentrates in expensive boroughs because the residual sustains the fees. Sophisticated consultants run scheme designs through their portfolio of borough-specific officer preferences before submission; they negotiate scope in pre-app; they manage withdrawal-and-resubmission strategies that typically low-volume independent applicants do not. The same scheme, prosecuted by a top-tier London planning consultant with a track record at the borough, has a materially different probability of approval than the same scheme prosecuted by an owner-architect who has never worked the borough's character appraisals.

This effect produces the same correlation between price and approval rate that the regressive-cost-asymmetry mechanism produces, and the same correlation that the decision-culture mechanism produces. It is observationally equivalent at the cross-sectional level. The LSSPD as currently constructed records the applicant name but not the planning consultant or the agent, so we cannot directly estimate the repeat-player effect. A clean econometric treatment would require agent-level identifiers across boroughs, which the dataset does not carry.

What the dataset does carry, indirectly, is some evidence that the agent-quality effect cannot fully exhaust the cross-section. The Bexley–Havering pair sits within the same applicant-pool quality distribution and still produces a 40pp gap. The Croydon natural experiment sits within the same applicant pool across the pre/post boundary (see §7) and still produces a 23.5pp shift. Neither is a complete rejection of the agent-quality hypothesis; each is consistent with agent quality contributing some share of the cross-sectional differential, but not the whole of it.

The regressive economics of preparation

A £60k preparation spend is in the order of 1% of a Westminster scheme's residual, roughly 11% of a Havering scheme's and over 20% of a Croydon one. On simple ROI grounds, applicants in more affordable boroughs have reason to underinvest in preparation relative to those in expensive ones. They are no less capable; the system simply makes preparation pay less there. This is the "regressive cost asymmetry" reading: a mechanism in which the visible cross-sectional pattern is produced upstream of any decision, by a filter that screens applicants on their ability to absorb preparation costs as a fraction of residual.

Two pieces of internal evidence are consistent with this mechanism operating in the visible window. First, the rising share of "insufficient information" and "insufficient detail" refusals in affordable boroughs (see §6) is the direct fingerprint of under-prepared applications. Second, the withdrawal-rate asymmetry (also §6) is consistent with expensive-borough applicants having consultant intelligence that affordable-borough applicants lack. Neither is a clean test of the mechanism in isolation, but both are consistent with it.

Why neither pure mechanism exhausts the cross-section

First, the Bexley–Havering pair sits within the same applicant-pool quality distribution at identical price points and produces a 40pp gap. The agent-quality reading is weak here: Bexley sustains higher outcomes on observably similar inputs.

Second, the Croydon political-revocation case is the hardest piece of evidence to reconcile with a pure applicant-quality interpretation. Croydon's applicant pool and the agents serving it did not change on 25 July 2022. The same applicants, prosecuted by the same agents, facing the same residual economics on the same sites, were approved at 56.3% pre-revocation and at 32.8% post-revocation.

The most defensible reading is therefore that the cross-section is generated by multiple mechanisms in unknown proportions. A substantial share of the affordability correlation is likely the regressive economics of preparation and the agent-quality concentration described above. A further share is likely downstream borough decision posture. The relative weight of the two mechanisms is not recoverable from this dataset alone; it remains a hypothesis warranting further work rather than a finding this essay establishes.

A within-agent test

The confound noted above, that the same price-approval correlation is produced by both the decision-culture and the applicant-quality mechanisms, can be partially tested by holding agent identity constant across boroughs. If the differential were entirely an artefact of agent sorting (better agents concentrating in permissive, expensive boroughs), then fixing the agent and varying the borough should reduce the estimated borough effect. Conversely, if the differential survives once you control for who the agent is, decision-culture is doing independent work.

We identified 29 named planning and design agents with ≥5 applications across ≥2 boroughs in the LSSPD (199 applications; 13 boroughs). On this sub-sample we ran two logistic regressions: one with agent fixed effects, one without, on identical observations and covariates (site type, conservation, PTAL band, unit count).

The results are unambiguous on rank-ordering but mixed on magnitude. Borough rankings are perfectly stable: Spearman ρ = 1.0 between the with-FE and without-FE borough coefficient; the same boroughs come out hardest and easiest regardless of whether agent identity is controlled. Individual coefficients shift (H&F, Tower Hamlets, and Southwark all have larger raw differentials than within-agent differentials, consistent with better agents concentrating there), but the rank ordering does not. A non-parametric check reinforces this reading: 79% of qualifying multi-borough agents show >10pp different approval rates across the boroughs they themselves work in. An agent achieving 65% approval in one borough earns around 45% working with a similar brief across the boundary.

Two caveats bound the interpretation. First, the sub-sample is thin: 29 agents, 199 applications, 13 boroughs. The key hard boroughs (Havering, Croydon, Waltham Forest, Newham) have no qualifying multi-borough agents and are therefore absent from the within-agent test. Second, 63% of the actor dataset has no resolved agent identity; the selected named-agent sub-sample over-represents the sophisticated end of the market, which is the very segment the agent-quality reading most concerns.

Within those bounds, this is evidence. The borough ranking is not an artefact of which agents happen to work there. The same agent, working the same types of scheme, achieves systematically different outcomes across boroughs. That is directional evidence for decision-culture and against a pure applicant-quality reading. The essay's characterisation of "multiple mechanisms in unknown proportions" remains correct; these results shift the weighting of the available evidence toward the decision-culture mechanism.

The inverted agent premium. A related finding from the same actor dataset needs care. Named professional agents (planning firms, architecture practices) approve at roughly 50%, while applications with no named agent approve at 61%. The gap is a selection artefact: named agents disproportionately take on the harder schemes — contested backland, conservation areas, refusal-risk sites in the difficult outer boroughs — so the 11pp shortfall reflects scheme difficulty rather than agent underperformance. The correlation between the agent premium and the borough differential is r = −0.59 (p = 0.04): the premium is most negative in the most restrictive boroughs. This fits the applicant-quality mechanism that sophisticated agents concentrate in difficult boroughs, work uphill against structural restrictiveness, and still lose more often than unrepresented applicants prosecuting easier schemes. This warns against reading raw agent-approval rates as a quality signal.

Regressive outcome, not regressive intent. We use "regressive" throughout in the technical economic sense, describing the distributional shape of an outcome rather than the intent of any borough or official. A borough responding to local preferences expressed through its democratic processes can produce a city-wide regressive outcome without anyone acting in bad faith: the diagnostic question is about effects rather than motives. We do not claim, and the data does not support, that any borough is deliberately producing a regressive outcome.

How the refusal-reason composition varies

Across the London-wide dataset, design (DES) refusals account for 54.8% of coded first refusal reasons, which is the largest single category in the regime. They dominate in many hard and permissive boroughs alike (Croydon 78.1%, Camden 78.2%, Richmond 68.6%, Havering 55.6%, K&C 52.9%), with one notable exception: Newham, among the hardest boroughs, refuses far more on policy and space standards (design just 26.4%). Design is the most discretionary category in the policy toolkit, the most subjective, and the hardest to adjudicate on appeal. This is a signature of small-sites planning across the board rather than of the hard boroughs in particular.

What would falsify this reading. The interpretation advanced here would be materially weakened if any of several conditions held, each testable with data the dataset does not yet fully carry. A fuller agent-level control set might eliminate most of the borough variation that survives the present 199-application within-agent panel. Appeal-adjusted outcomes might converge once overturn rates were netted out. Longitudinal borough trajectories might show no persistence from year to year. Or small-sites delivery, once measurable on a like-for-like basis, might prove unrelated to the differential. None of these is settled here, and stating them is the point: this is an account that could be shown wrong, and these are the ways it would be.

§6. Two forces, one outcome

The 2022–2026 cohort lets us look at two things at once: where the system sits at a moment, and where it is moving. The trend constrains the diagnosis of the cross-section directly; the cross-section is regressive, the trend is converging, and the relationship between the two shapes what policy is possible.

What the longitudinal evidence can and cannot do. The strongest possible version of this essay would track every borough's differential year by year and let each borough serve as its own control. A 2019–2022 backfill from the GLA Planning London Datahub (PLD; 43,667 decided residential applications across 32 boroughs) now provides a seven-year panel when merged with the 2022–2026 LSSPD. The pre/post borough approval correlation is r = +0.607 (2019–20 vs 2023–25), so the rank-ordering is structurally stable. Croydon's full trajectory is visible: 77.2% (2019) → 62.1% (2020) → 53.6% (2021) → 51.8% (2022) → 24.6% (2023) → 41.7% (2025), with the decline well advanced before formal revocation. The 2021 London Plan adoption produced a modest effect (pre-Plan 70.1% → post-Plan 63.5%, −6.6pp), consistent with the Plan having reduced its small-sites ambition before adoption. Per-borough trajectories are now more credible but still noisy, and should be read as directional. One coverage caveat: the PLD backfill uses a broader description-match filter than the tight 1–9-unit LSSPD, so level comparisons across cohorts need caution, while trajectory and rank-correlation analysis remain robust.

The dynamic trend

Over 2023, 2024 and 2025 the wealth–approval correlation has weakened from +0.66 in 2023 to +0.55 in 2025, and the spread between the four hardest and five most permissive boroughs has narrowed from roughly 55 percentage points to roughly 43.

Year by year

The convergence is a London-wide average; underneath it, individual boroughs move at different speeds. Computing the differential separately for each year (2023 to 2025, the window over which the model's covariates are available) shows a system that is moderately persistent rather than fixed. The hard and permissive poles are stable: Havering sits near −22pp every year, Richmond near +20pp. But the middle reshuffles, and several boroughs make large multi-year moves. For example, Sutton swings from −10pp to +26pp, Enfield from −7pp to +16pp, while Croydon climbs steadily off a −30pp low. Enfield's trajectory has a plausible political explanation: Labour gained control of the borough in the May 2022 elections (replacing a hung council), and the new administration submitted a new Local Plan for examination in August 2024. That move typically signals a borough committed to demonstrating housing delivery and prepared to defend refusals at appeal. That combination is consistent with officers shifting toward approval on borderline decisions. Sutton is harder to attribute to a single event: political control has been stable under the Liberal Democrats since 1986, and no new local plan was adopted in this window. The most likely candidate is national policy pressure: Sutton's 2018 local plan was approaching the end of its currency as the December 2024 NPPF revision engaged the tilted-balance presumption for boroughs not on track to deliver against the standard method. Both swings warrant caution. The annual differentials are computed on thinner within-year samples than the full-cohort cross-section, and the year-to-year rank correlation slides to 0.59 by 2025. Individual borough trajectories should therefore be read as directional signals rather than precise estimates. The year-to-year rank correlation falls with distance (Spearman 0.76 between 2023 and 2024, 0.59 between 2023 and 2025), so the differential is a usable borough signal but not a frozen rank. The metric does not extend below 2023: the earlier GLA series carries raw approval only, with no covariates to build a mix-expected rate, and a coverage break at 2022→2023 makes a direct level comparison unsafe — the raw 2019–2022 trajectory is the one shown in the Croydon panel above.

The candidate structural mechanism in operation

When the political lift abates, the candidate mechanism is what remains. Two pieces of evidence are consistent with the regressive cost asymmetry operating beneath the headline convergence.

First, the rising share of "insufficient information" refusals in affordable boroughs. Refusals citing insufficient or missing supporting information have risen from 4.4% to 5.7% (refused base) in the affordable cohort. They remain lower in expensive boroughs, but rise there too, from 0.0% to 2.2%; the gap is in the level and the pace, not the direction.

Second, the withdrawal-rate asymmetry. Across the visible window, expensive-borough applicants withdraw applications at roughly twice the rate of affordable-borough applicants (12.2% versus 6.9% over 2023–2024). On a naïve reading, that makes expensive boroughs look more punitive. But the mechanism runs the other way. Expensive-borough applicants have planning consultants who read officer feedback and withdraw pre-emptively rather than push through to a formal refusal. Affordable-borough applicants, without that intelligence layer, take the refusal. The visible refusal rate in affordable boroughs is therefore too generous: it understates the gatekeeping, because the marginal case that disappears in Westminster is pushed through to a recorded refusal in Havering. The headline approval rates this essay analyses are themselves conditional on a filtered subset of cases (5.6% of all applications are withdrawn and excluded), and the filtration is asymmetric in the direction that makes the published cross-section a lower bound on the true outcome differential.

What this means for the policy frame

The right policy frame holds two things together. Some affordable boroughs do, on the cross-section and on the Bexley comparison, produce lower outcomes than their caseload mix predicts. And a steadily lengthening list of central compliance requirements is itself a plausible regressive cost asymmetry. Both are likely operating, in unknown proportions, and the relative weights are not yet recoverable from the dataset alone.

The political-economy implication. Hard-borough councils cannot defend their outcome patterns by pointing only to the cost asymmetry, because Bexley and Sutton sit at the same point on the affordability map and produce different outcomes. Central government cannot defend a steadily lengthening list of compliance requirements as housing-neutral, because each additional requirement is a candidate mechanism for the regressive cost asymmetry. The cumulative effect, if the mechanism is operating, is substantial and the boroughs least able to absorb it are the boroughs the same regulatory chain is asking to deliver the additional homes.

§7. The decisive test: local plans against the London Plan

The outcome differential tells you that a borough produces different outcomes; it does not tell you why. To close the gap between symptom and cause, the differential ranking must be laid alongside each borough's local plan and read against the London Plan.

7.1 What the London Plan actually says

The 2019 draft London Plan contained, in draft Policy H2, a presumption in favour of small housing developments of 1–25 homes in PTALs 3–6 or within 800 metres of a station or town centre, with a draft ten-year small-sites target of 245,730 homes. That presumption was deleted on the Inspectors' Panel's recommendation. By the time the Plan was adopted on 2 March 2021, the target had been cut to 119,250 and the operative wording reduced to a duty for boroughs to pro-actively support well-designed new homes on small sites:

"Boroughs should pro-actively support well-designed new homes on small sites (below 0.25 hectares in size) through both planning decisions and plan-making…" — London Plan 2021, Policy H2 A.

7.2 Two categories of difficult borough

Category one: the differential is consistent with adopted policy. In some boroughs the lower outcomes are legible in the development plan itself, meaning, restrictive density ranges, hard limits on backland and garden development. Where this is the case, the differential is lawful and at least partly anticipated by the plan; the question becomes a democratic one mediated through s.24(7).

Category two: the differential is not consistent with adopted policy. The local plan is broadly consistent with the London Plan and the borough still produces a materially lower outcome than its caseload mix predicts. Here the differential is not obviously explained by the development plan itself. The remaining variation may arise from administrative practice, applicant composition, political expectations, local interpretation of policy, or other unobserved factors — the data locates the gap, it does not name its cause.

A note on local democratic legitimacy

Outer-borough preferences for restriction on dispersed intensification are genuine, are democratically expressed in council compositions, and are not in themselves illegitimate. A council pursuing policies its electorate has voted for, within the strategic framework of the London Plan and the legal framework of the planning Acts, is exercising the local discretion the planning system was designed to permit. This essay's policy framing therefore does not call for borough preferences to be overridden because they produce regressive outcomes. It asks that those preferences be transparent about their distributional consequences, with the Mayor's s.24(7) conformity opinion at plan examination as the constitutional mechanism for mediating between local preference and strategic policy.

7.3 Croydon: a closely-watched political shift

Croydon is the single best documented example in London of how rapidly a borough's small-sites outcome pattern can move. It is not, strictly, a closed natural experiment. Treating each decision as independent overstates the effective sample size given clustering, and the approval rate was already trending before the change. But it is the closest thing to a natural experiment available, and it is consistent with political instruction shaping outcomes at large magnitude.

The political timeline

SPD2 adopted April 2019. Governance referendum October 2021. Conservative Jason Perry won the May 2022 mayoral election on a manifesto pledge to revoke SPD2. Full Council revoked it on 25 July 2022. The permissive era, however, began well before the design guide. Drawing on Croydon's own monitoring, Russell Curtis (RCKa, "Come Back to Croydon", 31 May 2026) shows small-site approvals climbing from 307 homes in 2015 to 555 in 2016 and over 1,200 in 2017 — peaking with nearly 400 homes approved in the final quarter of 2017, almost a year before the SPD was first published in draft — before falling to 1,005 in 2018 and 898 in 2019 as the SPD's policies took effect. Curtis attributes the surge to "a culture of permissiveness that stemmed from the Labour administration's re-election in 2014", not to the design guide; on his reading the SPD arrived after the peak and coincided with the decline.

The before-and-after numbers

Pre-revocation cohort (GLA Datahub, 2019 – Jul 2022): 2,087 decided, 56.3% approval. Post-revocation cohort (LSSPD, Feb 2023 – Mar 2026): 878 decided, 32.8% approval. A headline fall of 23.5pp should be read as a contrast between two multi-year period averages, not as the discontinuity at revocation. The pre-period approval rate was already falling steeply: 69.7% (2019) → 44.2% (the months immediately before revocation) → 24.6% (2023) → 41.7% (2025). Measured against the immediate-pre level of 44.2%, the step at revocation is closer to twenty points; measured against the 2017 approval peak that Curtis documents (see timeline above), the decline began years earlier still. The assessment is that Croydon's approval rate swung by roughly twenty points across a political cycle, with the formal revocation consolidating a trajectory already well in train.

A definitional caveat specific to how this essay counts. The LSSPD tracks the 1–9 unit tier, which is not identical to the London Plan's spatial small-site definition (sites below 0.25 hectares). Schemes that are physically small but propose ten or more homes fall outside our unit-count filter — Curtis's example is the December 2017 Purley approval (Russell Hill) that replaced three houses with thirty flats at 364 habitable rooms per hectare. Our Croydon series therefore under-captures exactly the small-plot demolish-and-rebuild activity that drove much of the borough's headline delivery, and Curtis raises the same definitional gap against Centre for Cities / GLA analyses that counted only minor (sub-10-home) applications.

The within-period trajectory adds nuance: in 2019 Croydon's approval rate sat roughly 15pp above its mix-expected level (consistent with a permissive political culture that was, by then, already two years old and approvals had peaked in 2017, before the SPD); through 2020–2021 the rate fell as political resistance built; the post-revocation period produced outcomes around 20pp below mix-expected (consistent with political opposition in the restrictive direction). The mechanism of political culture instructing how officers read "character" is consistent with both endpoints of the trajectory, and with the fact that the permissive era both predated and the restrictive era postdated the formal instrument. It is the culture, not the design guide, that the data tracks.

7.4 Havering and Newham: the policy and the politics

Havering's posture is overtly defended in policy. Its Matter 20 hearing statement to the London Plan Examination in Public (EiP) argued that proposed suburban intensification was "ambitious to the point of absurdity," and the adopted 2016–2031 Local Plan concentrates growth on Romford and Rainham/Beam Park.

Newham's 2018 Local Plan remains the adopted plan; a new Local Plan that strengthens family-housing protection is at examination (submitted July 2025) and not yet adopted. The emerging plan retains the strategic-sites framing while keeping discretionary character protection across the wider suburban tier.

The two boroughs make that case with different strength. Newham's restraint maps onto a specific, defensible housing-mix concern, protecting family-sized homes from subdivision into flats, which a development plan can legitimately own. Havering's is harder to locate in the plan itself: its adopted policies are largely the ordinary character and design tests, so how far its lower outcomes are plan-consistent rather than a matter of discretionary practice is the question §7.2 leaves open, and one its appeal record can test. Where a borough's restraint is genuinely plan-led, the constitutional mechanism for mediating between local choice and city-wide strategic policy is the s.24(7) conformity opinion at plan examination.

7.5 Westminster and Richmond: high approval as a post-filter outcome

Westminster's small-site caseload is composed of applicants who can absorb £60–80k of preparation cost as a single-digit-percentage drag on residual. Those applicants arrive already prosecuted by consultants who have read the borough's policy carefully. Westminster's high approval rate is therefore, at least in part, a post-filter rate. A policy edit could not transfer that pattern to Havering, because the upstream filter that produces Westminster's caseload is driven partly by Westminster's property values rather than by its planning department.

Running the test requires three inputs, all publicly available or recoverable from the dataset. First, the differential itself, per borough, from this dataset. Second, the borough's adopted development plan policies on small sites: specifically, whether they contain explicit restrictions on density, backland or garden development, building heights or plot ratios that go materially beyond the London Plan's H2 duty to pro-actively support well-designed small homes. These policies are public documents and typically short. Third, the coded refusal reasons for the borough; which policy grounds appear most frequently, and whether those grounds are the ones the development plan would predict, or whether they cluster around character, design, or precedent grounds that are harder to trace to any specific written policy.

A positive result (Category 1) looks like this: the borough has a materially restrictive density or character policy; officer reports cite that policy consistently; and Inspectors broadly uphold the refusals on appeal. The differential is lawful, and the constitutional route is plan examination via s.24(7). A negative result (Category 2) looks like this: the development plan has no unusually restrictive small-site policies; refusals nonetheless cite character and design grounds at a rate well above the London average; and a disproportionate share of those refusals is overturned on appeal. The data for this test exists for every borough in this dataset: refusal reason coding is in the LSSPD cohort, and Planning Inspectorate (PINS) appeal outcomes are cross-referenced by borough and decision date. A borough in Category 2 does not necessarily have a legal problem, but it has a transparency obligation and an appeal-cost liability. This would be evidence that its decisions are diverging not just from the London Plan, but from the Inspectors' reading of its own adopted policies.

§8. Implications for Local Authorities and for central government

This section is addressed to two readers in turn: first the authority reading its own number, then the government deciding what, if anything, to do about the pattern. The order is deliberate. The differential is built from each borough's own decisions, and the first claim on its meaning belongs to the borough.

8.1 Three readings of a negative differential

A borough sitting below expectation has at least three accounts available to it. They are not mutually exclusive, the data in this essay cannot apportion them for any individual Authority, and each carries a different implication but each is testable against evidence the borough already holds.

The first is deliberate, plan-led restraint. A borough whose adopted plan expresses a settled, democratically endorsed position on dispersed intensification and which applies that position consistently will sit below the London average by choice. That is not the development management function failing but rather the function doing what the plan instructs. The test of this reading is coherence, and §7 sets it out: refusals that cite the plan's policies accurately, fall consistently on the typologies the plan constrains, and are broadly upheld by Inspectors on appeal. A borough that passes that test holds a Category 1 position: a lawful, defensible differential whose proper forum is plan examination, not the development management statistics.

The second is caseload quality on the dimensions the model cannot see. The differential controls for four observable characteristics; it is blind to preparation. A borough receiving systematically less-prepared applications (thinner supporting information, less pre-application engagement, less experienced agents) will sit below expectation without deciding any harder than its peers. The test of this reading is in the refusal record: a rising share of insufficient-information refusals, validation failures, and schemes lost on matters a competent submission would have resolved before it arrived.

The third is capacity. A development management service under sustained resourcing pressure can drift below expectation without any policy choice at all: less pre-application capacity to repair weak schemes before submission, less officer time to negotiate amendments rather than refuse, more decisions taken on the papers as received. The test here is procedural: determination times, amendment rates, the balance of negotiated approvals against first-pass refusals.

An Authority that can say which of these accounts, in what proportion, explains its own number has already met the standard §11 proposes (explained variation) and met it on its own terms.

8.2 What the differential offers a borough

Used internally, the differential is management information of a kind no published statistic currently provides: a mix-adjusted account of the authority's own decision record, benchmarked against what its caseload would have produced under London-average behaviour. Four uses follow directly. In the Authority Monitoring Report, it gives the service a like-for-like measure to sit alongside the raw approval rate and a defensible answer to the member, journalist or Inspector who quotes the raw rate without the caseload behind it. In local plan evidence, the typology-level patterns beneath the headline show where the adopted plan is being applied as intended and where outcomes have drifted from the position the plan expresses. This signals evidence of consistency for examination, or early warning of a gap between the plan as written and the plan as applied. In member briefings, the three readings above give officers a vocabulary for the borough's position that neither concedes failure nor refuses the question. And at appeal, an Authority whose refusals are consistent with its differential, its plan and its Inspectorate record can evidence a coherent decision posture rather than defending each case as though it stood alone.

8.3 The finding that protects the restrictive boroughs

The most protective finding in this dataset belongs to the hardest boroughs: their refusals are upheld on appeal. Across the matched appeal cohort (§9.6), the most restrictive boroughs are overturned at roughly sixteen per cent of decided appeals, which is below the permissive boroughs' twenty-four, and below the national minor-residential rate. Whatever else a persistent negative differential means, in these boroughs it does not mean capricious decision-making that the Inspectorate corrects on review; it is consistent with a coherent position, consistently applied, that survives independent scrutiny. An Authority confident in the first reading above will find its strongest evidence here, and is entitled to use it. The usual caveat travels with the number: only around six per cent of refusals are appealed, so the sample is selected by applicant judgement but the cases that reach the Inspectorate are those applicants judged most vulnerable to challenge, which makes the upheld share harder, not easier, to dismiss.

8.4 The burden runs both ways

The cost asymmetry this essay describes is usually read from the applicant's side: preparation pays least where values are lowest, so the least-prepared applications concentrate in the most affordable boroughs. Read from the Authority's side of the counter, the same mechanism is a workload statement. The development management teams serving the most affordable boroughs are processing a rising share of applications that arrive unready; the insufficient-information trend of §6 is its visible edge. That work carries officer hours, a negotiation burden and appeal exposure, typically on thinner resourcing than their inner-London counterparts. An Authority in this position is not the author of the asymmetry but one of the parties carrying its cost. A policy response that treats the asymmetry solely as an applicant grievance has misread where the burden falls — which is why the proportionality test proposed at 8.11 is as much a protection for the boroughs asked to administer each new requirement as for the applicants asked to meet it.

8.5 Access to the underlying evidence

The number in this essay should not be the last word on any authority. Any London borough wishing to interrogate its own differential may request its case-level extract and the full methodology annexe, without charge. The differential is offered as the beginning of an account each authority can give in its own terms, on its own evidence, not as a verdict delivered from outside it.

The remainder of this section turns from the authority reading its own number to central government and the GLA deciding what, if anything, the pattern asks of them. For that reader, the differential works as a diagnostic rather than an accusation, and the policy frame it points to is a transparency regime rather than a central-override one.

8.6 Treat the outcome differential as a published transparency instrument

Calculate annually, publish, and read alongside local plans. Its function is as a standing question put to each authority: your caseload predicted X; you produced Y; account for the difference. The point of publication is not to shame boroughs but to give the democratic process the information it needs to mediate between local preference and strategic policy.

To make that concrete: publish the differential each year alongside the borough's small-sites monitoring, and ask each authority to attach a one-page note answering three questions. (1) Which quadrant of the delivery matrix are we in — what is our small-sites posture, and what is our delivery record? (2) What explains our position — caseload, viability, land supply, decision posture, or some mix? (3) What, if anything, are we doing to change it? This is deliberately not a performance league table, and it should not be administered as one: a borough in the "restrictive and productive" quadrant may give a complete and creditable answer. It is a standing prompt for democratic accountability, so that a borough's own voters can see the trade-off it is making between local character and city-wide supply, and judge it on terms the borough itself has set out.

8.7 Use the Mayor's s.24(7) conformity opinion as the principal lever

The Mayor's call-in and direction-to-refuse powers, expanded by Potential Strategic Importance (PSI) Category 3J, cover 50+ home schemes; the 50-home floor deliberately excludes the 1–9 unit tier this essay tracks. The s.24(7) conformity opinion is the appropriate instrument where a borough's adopted plan has materially diverged from London Plan Policies H2 and D3. It is not, on this evidence, an instrument that should be used as a routine override of borough discretion in the absence of policy divergence.

8.8 Let appeal outcomes carry their full evidential weight

If a meaningful share of the differential is officer or member risk-aversion, the cost of unjustified refusal must rise on appeal. The practical reform is not legislative but it is aimed at transparency. Government should require boroughs to publish, alongside their annual monitoring reports, the proportion of their small-site refusals overturned at appeal and the costs awarded against them.

8.9 Reframe affordability targets around the supply geography

This is a policy judgement, not a finding the data dictates. If raising housing supply in the most affordable boroughs is a policy objective, as capital-wide affordability ambitions imply, then persistent negative differentials in exactly those boroughs deserve particular scrutiny: on this evidence, the places where homes are most affordable to deliver are among those where they are hardest to consent relative to caseload. Where a borough's local plan has explicitly chosen that outcome, the mediation runs through s.24(7); where it has not, the gap is at least a question the borough should be asked to answer. The data establishes the gap; whether it should concern us is a judgement about what the planning system is for.

8.10 Learn from the affordable-and-permissive boroughs

Bexley, Sutton, and Enfield produce constructive small-sites outcomes at the most affordable end of the market. The Bexley–Havering forty-point gap at identical price points is the single most actionable data point. It should be read as evidence that the cross-section is not destiny, rather than as grounds for criticising Havering.

8.11 Act on the regressive cost asymmetry

Using Land Registry transaction prices, the preparation cost burden on a 5-unit scheme already varies roughly twenty-fold across London with ~1% of residual in K&C and Westminster to ~21% in Croydon. It is now a measured gradient rather than a mechanism still to be tested. Government should commission a cumulative-cost impact assessment of small-site validation requirements, expressed as a percentage of expected residual value across borough price quartiles, and publish it. The regressive outcome of additional validation requirements would then be established.

The same logic should run forward as well as back. Every new validation mandate on small sites, biodiversity net gain is the live example, should pass a small-sites proportionality test before adoption: model the preparation cost the mandate imposes as a percentage of expected residual value across borough price quartiles, and check whether it screens out a materially larger share of would-be applicants at the affordable end than at the expensive end. A requirement that costs a Westminster scheme a deductible afternoon and a Havering scheme a fifth of its residual is not neutral, however neutral it looks on the face of the regulation. The test does not pre-empt the policy; it makes the distributional shape visible before, rather than after, it is imposed.

§9. Implications for developers

9.1 The borough is an acquisition input

The differential implies that the identity of the determining authority is a first-order driver of consent outcome, partly independent of how good the scheme is, though the relative contribution of borough decision posture and applicant-quality filtering on that outcome is unresolved. Either way, the borough variable is key in acquisition appraisal as a risk premium.

9.2 Read your borough's quadrant before fixing the brief

The delivery matrix tells you which question to ask of a target borough. It does not tell you what to bid, and it cannot: the delivery axis is all-sites and lags the decision window, so a quadrant is a starting question about where risk concentrates, not a site-level approval probability. Read it that way.

A quadrant is a borough-level starting point, not a site-level answer. The same cell can be produced by different mechanisms, and the matrix's delivery axis is all-sites and time-lagged. Before relying on any borough cell, drop to the area- and decision-level reads below, which carry far more signal for a specific site than the borough average.

9.3 Do not read the ranking as a static verdict

Croydon's outcome pattern swung by twenty-one percentage points across a single political cycle. Treat the ranking as the current state of play, refreshed annually, not as a permanent property of a borough. Only the extremes carry confident signal.

9.4 In hard boroughs, the local plan is the brief

The same £60k of preparation that is a tax-deductible nuisance in a Westminster scheme is a material economic constraint on a Havering scheme. The temptation to economise on preparation in affordable boroughs is rational on per-deal economics and, if the cost-asymmetry mechanism is operating, contributes to the cross-sectional pattern this essay describes. The marginal benefit of being one of the well-prepared applicants in a hard borough is large precisely because most other applicants are not.

9.5 The gradient is not fixed

The correlation between affordability and below-expectation outcomes is strong, but it is a pattern, not a law, and the exceptions matter more than the rule. Bexley, and on the recent trend, Sutton and Enfield combine modest land values with outcomes above what their caseload mix predicts. For an applicant the implication is straightforward: these are the environments where consent risk is lowest relative to entry cost, and an appraisal must identify this. For the argument of this essay there is a larger implication: the existence of affordable boroughs producing above-expectation outcomes demonstrates that affordability does not oblige an authority to sit below expectation. The Bexley–Havering gap at near-identical price points is the most instructive comparison in the dataset, and the productive question it puts to any outer borough is an enquiry, not an accusation: what is Bexley doing that we are not?

9.6 Treat the appeal as a narrow instrument

PINS national allowed rates of approximately 24% for minor residential⁴ mean an appeal has a realistic prospect of success where a refusal rests on discretionary character reasoning unsupported by evidence. London's observed small-site overturn rate (2022–2026) is 21.0%. The pattern varies by borough: hard boroughs overturn at ~16%, permissive boroughs at ~24% (r = +0.40 between difficulty gap and overturn rate). Hard boroughs' refusals are upheld more often than not. This likely means they are applying a coherent if restrictive policy position rather than making capricious decisions that PINS would correct. An appeal is best used as a targeted instrument against a specific reasoning weakness, and is not a reliable means of reversing a marginal refusal. Caveat: ~6% of refusals are appealed, so the sample is selected.

9.7 Understand how the borough decides

In hard boroughs, outcomes vary within the authority as well as between authorities, and part of that variance sits in how decisions are organised: which applications are delegated and which reach committee, how pre-application advice is resourced, and how consistently the borough's published position is applied across comparable cases. Before you submit, understand the borough's decision-making structure — its delegation scheme, its committee's recent record on the typology, the pre-application route and what it has produced for schemes like yours — rather than relying on the borough average alone. Decision-level variation below the borough average exists in the data, but this essay deliberately aggregates it. Individual decision-makers are not its unit of analysis. Any record read at that grain risks being drawn disproportionately from refused cases, which would bias the picture in any event. The borough's structure and its recent record on comparable schemes carry the signal an applicant can responsibly use.

The practical yield, in one place. The actionable content of this analysis is deliberately modest, because the mechanism is unresolved, but it is concrete. For a developer it changes two operational decisions: which boroughs to weight in acquisition, and where to spend preparation effort and scrutiny once a site is in play. For a local authority it provides a mirror; a measured account of what its chosen posture is producing, framed as a question it can answer on its own terms. For central government it marks where transparency and a single proportionality test could remove unnecessary, distributionally-skewed cost without overriding a legitimate local choice. None of this tells any borough it must become more permissive. The whole of it tells every borough and every department that imposes a cost on it, that it should be able to say what its chosen level of permissiveness, and the burden behind it, is actually achieving.

§10. What this means for housebuilding

Step back from the boroughs and the question becomes both national and international. Small sites are 23% of London's ten-year housing target by the GLA's own reckoning. This evidence is consistent with a meaningful fraction of London's small-site capacity being affected by two intertwined dynamics: a borough decision regime whose outcomes vary by roughly forty-three percentage points across the capital, and a regressive compliance-cost asymmetry (preparation cost ~1–21% of residual across the city, r = −0.47 with the outcome differential). Both fall hardest on the boroughs where homes are most affordable.

One clarification derived from the national data: this is London's problem, not England's. Replicating the analysis against ~280 English LPAs and the data paints a different picture. Whatever pathology the London data captures, it is not a universal consequence of English discretionary planning. The reforms in §8 are London-specific.

What international evidence says about discretion itself. Four jurisdictions that shifted from discretionary to rules-based entitlements offer a framing lens. Auckland's 2016 Unitary Plan upzoning covered ~three-quarters of residential land in a single instrument. The result was an estimated ~22,000 additional dwelling consents over five years, with rents estimated 14–35% lower than the counterfactual (a figure since contested).⁵ Minneapolis 2040 (2020) grew housing stock 12% while rents rose 1%, against 4% and 14% for the rest of Minnesota; the mechanism was by-right large multifamily near transit, with the headline triplex allowance making up only ~1% of permits.⁶ California's accessory dwelling unit (ADU) laws stripped local discretion on these units and drove permitting from under 5,000 (2017) to over 23,000 (2022), roughly one in five of all new California permits, while the nominally similar SB9 (duplex by-right, 2022) produced negligible uptake where friction remained.⁷ Japan's ~12 standard national use-zones remove case-by-case discretion entirely: roughly 940,000 housing starts annually against ~194,000 in the UK.⁸ The common thread is the mechanism rather than the density permitted: where capacity becomes a predictable entitlement, supply responds; where the outcome is still case-by-case, the formal permission delivers little. England's small-site regime sits in the second group, and the wide variation documented in this essay is the direct measure of that negotiation's price.

The implication for housebuilding is both sobering and uncertain. Sobering, because the cross-section pattern is well-evidenced, descriptively skewed by price, and visible in twelve thousand decisions. Uncertain, because the mechanism producing the pattern is not identified by this data; both the decision-culture reading and the applicant-quality reading are consistent with what we observe.

Croydon shows quantitatively that a borough's outcome pattern can move by approximately twenty-one percentage points across a single political cycle. The Bexley–Sutton–Enfield counter-cases suggest that constructive outcomes at the affordable end of the market are reachable. The 2023–2025 post-NPPF convergence shows that central political pressure can lift the visible approval gradient by eight percentage points at the affordable end of the market when applied. The partial reversion already visible in 2026 suggests the political-pressure lever has a limited shelf life. The structural reforms (borough plan-making where it has materially diverged from strategic policy; central compliance cost where it has accumulated without distributional assessment) are what would outlast the politics.

The geography of "no" can change. It is a set of outcomes, generated borough by borough by mechanisms that this data partially identifies and partially does not, against a regulatory backdrop constructed instrument by instrument without assessing its cumulative distributional shape. The first step is to stop comparing boroughs by raw approval rates that flatter some and malign others, and to start asking each authority a fair question that admits multiple legitimate answers: given what you received, what outcome did you produce — and where that outcome differs materially from elsewhere, what public benefit justified the difference? The differential does not deliver a verdict; it makes that question answerable. The second step is to ask the same question of central government: given the compliance burden you have imposed, what share of the country's would-be small-site applicants can afford to clear it, and where do they live? Neither step requires anyone to act in bad faith. The current state, in which the answer to both questions is "we don't know", is the only state that is genuinely indefensible.

§11. What level of variation is acceptable?

The preceding sections establish that variation exists, that it is large, and that its mechanisms are only partly identified. They do not engage the harder question, which no enlargement of the dataset will answer: whether, and at what point, that variation is acceptable. This is a normative question, not an empirical one, but it is the question the differential is ultimately in service of, and it is set out here.

Variation is not, in itself, a defect. A planning system engineered to produce identical outcomes in every borough would have abolished the local discretion the planning Acts were written to create. Some divergence between places is not a fault to be corrected; it is the visible price of letting places decide for themselves, and a London in which every borough consented small sites at an identical rate would be a London that had stopped being governed locally. The question is never "why is there variation?" but "how much, and on what account?"

Scale is what turns a feature into a problem. A forty-three-point spread in outcomes under like-for-like observable mix goes well beyond the fine grain of local character: it is large enough that whether a Londoner can build a home, and at what cost, depends heavily on which side of a borough boundary the site happens to fall. And the capital's housing requirement is a shared obligation apportioned to boroughs by the London Plan, not a set of thirty-two private budgets. When one borough's outcome runs twenty points below what its caseload predicts, the homes that go unbuilt there do not leave the city's need; they are displaced onto other boroughs, other tenures, or out of the market altogether. This is the asymmetrical outcome: discretion exercised at the scale of a borough has city-wide consequences that local discretion, by its very construction, is not built to weigh.

A borough's decisions are made by, and answerable to, the people who already live there. The cost of a refusal falls most heavily on people who do not. The larger and more persistent a borough's negative differential, the larger the cost carried silently by a constituency with no vote in the decision. This does not make the borough wrong; §7 defends local discretion without reservation. But it does mean the acceptable magnitude of variation is not a matter each borough can settle entirely on its own, because the people who bear the largest share of the cost are structurally absent from the room in which it is decided.

The workable standard, then, is explained variation. Boroughs are entitled to differ from the London average; the test is whether a borough can account for the difference in terms a reasonable observer, including the people the difference excludes, would accept. A borough whose adopted plan has openly chosen to constrain dispersed intensification is exercising a legitimate choice it can defend on the record. A borough producing the same outcome with no policy basis, no stated rationale, and no acknowledgement of the distributional consequence has not yet been held to account for it. The differential does not fix the acceptable level of variation. What it does is make the variation legible, so that the question of justification can be asked at all.

There is, in principle, machinery for adjudicating the conflict between local discretion and shared obligation and its limits are instructive. At plan-making, the Mayor's conformity opinion under s.24(7) tests a borough's local plan against the strategic policy of the London Plan; at the level of individual decisions, the transparency regime §8 proposes would ask boroughs to account for large differentials in the open. But both engage policy, not outcomes. The Mayor reviews what a plan says, not what a borough's decisions produce; and no existing instrument asks a borough to reconcile a persistent outcome differential with the city-wide need it is part of meeting. The system can mediate divergence that is written into a plan, but has no way to notice divergence that resides in the pattern of decisions. That is the gap a published differential is meant to fill not by overriding the borough, but by giving the people who bear the consequences, and the institutions that speak for the city, something measured to hold it to.

What level should actually concern us? There is no one-size-fits-all number, and providing one would be the same false precision the rest of this essay has worked to avoid. But the conjunction is recognisable. Variation that tracks an adopted local plan, or that sits within the statistical noise of a thin annual caseload, is unremarkable. The pattern that warrants concern is the compound one: variation that is large, persistent across years, unexplained by the development plan, and concentrated in the boroughs where homes are most affordable and the city's need is greatest. That conjunction proves nothing about cause but it is precisely the signature the cross-section shows, and precisely the signature the standing question in §8 is designed to surface.

So the acceptable level of variation is not a number to be computed but a judgement to be made and made, ideally, by people who can see what the variation is producing and on whose behalf it is produced. The differential cannot determine that judgement. It can only insist that the judgement be made in the open, against a measured quantity, rather than left to the accumulation of individually reasonable decisions whose collective shape is unforeseen. That is the limit of what measurement offers.

§12. The question the differential cannot answer

Suppose every confound in §5 were resolved: we could split the differential cleanly into borough decision posture and upstream applicant-quality filtering, and a borough's negative differential turned out to be, in measured part, a genuine choice to decide harder than its caseload predicts. The hardest question would still be open, because it is not an empirical one and no enlargement of the dataset will close it.

A borough's planning decisions are made by, and answerable to, the people who already live there. That is what local democratic control means, and §7 defends its legitimacy without reservation. But the people most affected by a refusal are often not the residents represented in the decision. They are the people who would have lived in the homes that were not built but priced into a different borough, a smaller home, a longer commute, or out of the market altogether. Their interest is as direct as any objecting neighbour's; they simply have no vote, because they do not yet live there, and many never will, precisely because the decision went the way it did. This is not a charge of bad faith; it is closer to the opposite. A council can weigh every represented interest scrupulously and still produce an outcome that is, at the scale of the city, distributionally regressive, because the constituency that bears the cost is absent from the room. The asymmetry is built into how local democracy is constituted, not introduced by any official within it.

The structural asymmetry, in one line. Existing residents vote on the decision; future residents do not. The metric may, in the end, be measuring the consequences of that gap.

The differential does not resolve this, and it would be a category error to ask: no statistic adjudicates a conflict between people who are represented and people who are not. What it can do is make the conflict visible by putting a number on the distance between the outcome a borough produces and the outcome its caseload predicts, and letting that number be read against the homes it implies were not consented, and the people they would have housed. Transparency does not remedy the representation gap; short of a remedy, it is the most honest response available.

That is the limit of what this essay claims. The metric measures variation. Whether a given level of variation is acceptable is a question about whom a planning system is built to serve. The contribution here is to take a debate usually conducted in anecdote and give it a measured quantity to begin a dialogue.