Working paper no. 2 · Preregistered 2026-06-11 · 151 local authorities · 2023 & 2024
The SEND Friction Index
Measuring the administrative burden on parents.
Six official measures of the fight — missed statutory deadlines, refusals to assess, refusals to issue, formal challenges, ombudsman complaints, overcrowded special schools — combined into one score per local authority. Friction is large, persistent, and not deprivation in disguise. And where it is highest, more families exit the system into home education the following year.
The map · friction quintile bands, 2024
Where the system fights its families hardest
Loading map…
High-friction authorities are everywhere — London boroughs, shire counties, northern cities. The score correlates only weakly (and negatively) with deprivation: this is a property of local administrative systems, not of poverty. An authority's band is also persistent: the 2023 and 2024 scores correlate at ρ = 0.72 across the 151 authorities scored.
What's in the score
Six failure modes, process only — outcomes deliberately kept out
Each component measures what the administrative system does to families, never how children fare — child outcomes are reserved for testing the index, not building it. Components are standardised within each year and averaged with equal weights; an authority needs at least 4 of 6 published components to be scored.
20-week timeliness failure
Share of EHC plans not issued within the statutory 20 weeks, exceptions included.
Refusal to assess
Requests for an EHC needs assessment refused, per decided request.
Refusal to issue
Decisions not to issue a plan after assessing, per concluded assessment.
Formal challenge volume
Mediations plus tribunal appeals against request decisions, per 10,000 plans. Total challenge, deliberately — Paper 1 showed the tribunal-only rate measures family resources, while the total is flat across deprivation.
Ombudsman complaints
LGSCO Education & Children's Services complaints received, per 10,000 plans.
Category is broader than SEND — includes admissions, transport and children's social care.
Special-school occupancy
Pupils on roll per commissioned place in state special schools.
Above 1.0 = over capacity. Some authorities deliberately run above number.
The spread · component means by friction quintile, 2024
What a top-band system looks like from the kitchen table
| Component | Q1 · least friction | Q5 · most friction |
|---|---|---|
| Plans missing the 20-week deadline | 31% | 59% |
| Assessment requests refused | 19% | 38% |
| Assessments ending in no plan | 2% | 10% |
| Formal challenges per 10k plans | 151 | 420 |
| LGSCO complaints per 10k plans | 67 | 122 |
| Special-school occupancy | 0.98 | 1.11 |
The components travel together loosely, not as one syndrome — the first principal component carries only 29% of their variance. Authorities that refuse a lot are largely not the ones that delay a lot (refusing fewer assessments leaves fewer to run late). The single score summarises average intensity; the component breakdown, published for every authority, is the diagnosis.
Exploratory correlates · labelled as such in the preregistration
Friction is not deprivation in disguise
IMD deprivation
−0.07
index SD per predictor SD · slightly lower where deprived
Official appeal rate
+0.17
index SD per predictor SD · strong convergent check
Authority size (log pupils)
+0.11
index SD per predictor SD · bigger systems, more friction
The index moves with the DfE's independently-constructed tribunal appeal rate (+0.17 per SD, p < .001) — places we call high-friction are the places families actually appeal. It is unrelated to Safety Valve deficit status, and barely related to deprivation — in the wrong direction for a poverty proxy. Whatever this measures, it is not the same thing as disadvantage.
Preregistered validation · index in year t → outcomes in t+1
The outcome friction moves is the exit door
We locked three predictions before looking: high-friction authorities should show wider persistent-absence gaps (V1), wider suspension gaps (V2) and more home education (V3) for the following year, with no-SEN absence as a negative control (V0). V1 and V2 failed. V3 held.
Home-education rate in t+1
Per 100 pupils, by friction quintile in year t. Conditional estimate: +0.135 per SD (p = .009), ≈ +11% of the mean rate.
Absence gap in t+1 — the null
The raw rise across quintiles is fully absorbed by need mix, deprivation and region — the conditional coefficient is −0.09 (p = .84). We predicted otherwise; the preregistration committed us to reporting this as a null, and it is one.
| Model | β | Robust SE | 95% CI | p | N |
|---|---|---|---|---|---|
| V1persistent-absence gap, EHC vs no-SEN (pp) | −0.086 | 0.436 | [−0.941, +0.768] | .843 | 297 |
| V2suspension-rate gap, EHC vs no-SEN (per 100) | +0.529 | 1.137 | [−1.700, +2.757] | .642 | 297 |
| V3elective home education rate (per 100) | +0.135 | 0.051 | [+0.034, +0.236] | .0088 | 289 |
| V0persistent absence, no-SEN pupils (negative control) | +0.132 | 0.224 | [−0.307, +0.570] | .556 | 297 |
The negative control behaves — friction does not predict absence among children with no SEN (p = .56) — but the preregistered composite criterion was conditioned on V1, so by its own terms the index is not validated as a predictor of within-school disadvantage gaps. What survives is sharper: on a one-year horizon, administrative friction predicts whether families stay in the system at all.
| Model | β | Robust SE | 95% CI | p | N |
|---|---|---|---|---|---|
| R1PCA-weighted index (V1) | +0.170 | 0.189 | [−0.201, +0.541] | .369 | 293 |
| R2Rank-based index (V1) | −0.089 | 0.203 | [−0.488, +0.309] | .661 | 297 |
| R4Exclude London (V1) | −0.097 | 0.509 | [−1.095, +0.901] | .849 | 234 |
| R5WLS by pupils (V1) | −0.192 | 0.448 | [−1.071, +0.687] | .668 | 297 |
| R6Friction × IMD interaction | +0.591 | 0.470 | [−0.330, +1.513] | .208 | 297 |
| R7a2023 transition only (V1) | +0.071 | 0.607 | [−1.118, +1.261] | .906 | 146 |
| R7b2024 transition only (V1) | −0.233 | 0.464 | [−1.143, +0.677] | .616 | 151 |
| R8Exclude inherited-IMD LAs (V1) | −0.233 | 0.456 | [−1.126, +0.661] | .610 | 281 |
Leave-one-component-out re-estimates of V1 span β = −0.37 to +0.09 — never significant. The V1 null is not an artefact of any single component, weighting scheme, region, or year.
What this index is not
Read the bands, not imaginary decimal places
Not a league table. We publish quintile bands and leave-one-component-out rank ranges, never precise ranks. The median authority's rank moves across a ~38-place range depending on which component you drop — bands are the honest resolution.
Not a measure of how children fare. By design. It measures the administrative fight, and our own preregistered tests show it does not (yet) predict school-outcome gaps one year on.
Not all SEND. The ombudsman component counts all Education & Children's Services complaints, and home education is recorded for all pupils — both caveats are disclosed wherever those numbers appear.
Not causal. Two years of index data give two t→t+1 transitions: pooled associations with controls, nothing stronger. The third year lands June 2026 and the pipeline ingests it with a one-line change.
What follows
For families
Your authority's band and component breakdown turn 'everything is a fight here' into specific, citable claims — chronic delay is a different fight from systematic refusal, and the breakdown says which one you're in.
For policy
Friction is measurable from data the DfE already publishes, stable year to year, and unrelated to deprivation or deficits. A regulator could compute this tomorrow and target inspection with it.
For research
The validated margin is exit: more home education follows high friction. Whether longer exposure widens in-school gaps is exactly what the accumulating panel will test from June 2026.
Methods & materials
Six components from five official sources (DfE SEN2, school census, SCAP, LGSCO datasheets, ONS boundaries), winsorized and z-scored within year, equal-weighted, ≥4 of 6 required. Validation models, robustness battery and interpretation rules locked before estimation; raw snapshots pinned by SHA-256; every number script-generated. Analysis executed by an LLM research agent under human direction — and designed so you don't have to take anyone's word for it.
See your local authority's component breakdown in the LA explorer, or read Paper 1: Refused, Not Appealed.