Headlines

  • Over the last 4-years, after accounting for structural factors, Brighton and Hove ranks (out of 152 LEAs):
    • 7th BEST for Disadvantaged GCSE Attainment
    • 5th BEST for Non-Disadvantaged GCSE Attainment
  • Last year, Brighton and Hove was 2nd WORST for absence.
  • GCSE Attainment is very predictable at School Level - 80% variation explained by a handful of variables
  • The biggest factor affecting disadvantaged student attainment is absence
  • Data shows disadvantaged students in England do better in schools with more disadvantaged students in
  • Poor understanding of evidence → poor policy → poor parental choices

Motivation & Context

Why this paper?

  • The UK Government’s 2026 Schools White Paper sets ambitious targets:

    • Halve the disadvantage attainment gap
    • Improve attendance
    • Reform admissions
    • Tackle place-based disadvantage

  • Achieving these requires LEAs to understand local drivers of attainment

  • This paper is motivated by direct experience of policy development in Brighton and Hove in 2024–25

The problem: A crucial gap exists between the availability of open data and the availability of accessible, contextualised intelligence for decision makers.

The risk: Policy designed around an incomplete understanding of local attainment drivers can pull the wrong lever — doing little good and potentially causing harm.

What are we trying to do?

  1. Show that DfE open data can explain ~80% of variation in school-level attainment with just a handful of variables

  2. Identify the relative importance of different policy levers — and reveal which ones actually have mechanical advantage

  3. Apply this to Brighton and Hove as a case study where policy pulled the wrong lever

  4. Present a Policy Simulator tool that bridges the gap between data and decision making

We are not claiming the best possible model — we are showing that good enough models built from open data can transform the quality of local policy conversations.

What drives attainment at GCSE? The literature

  • Attendance: strongest predictor — persistently absent pupils: 36% pass rate vs 84% for full attenders (DfE, 2022, 2025)

  • Prior attainment at KS2: reflects accumulated earlier inequalities (Gorard and Siddiqui, 2019; Stopforth and Gayle, 2025)

  • Disadvantage (FSM): some residual effect remains after accounting for above. Gap narrowed post-2011, reversed post-COVID (Tuckett et al., 2023)

  • Workforce: leadership quality, teacher retention, and teacher sickness matter — especially for disadvantaged pupils (Gibbons et al., 2018; Menzies, 2023; Zuccollo et al., 2023)

  • Selectivity: creams-off rather than improves - grammar school premium largely disappears after SES controls (Anders et al., 2024; Gorard et al., 2022; Gorard and Siddiqui, 2019)

  • Individual Pupil characteristics (Houtepen et al., 2020; O’Connell and Marks, 2022)

The factors are multifaceted, complex and interrelated. What’s needed is an analysis combining them so their relative importance can be assessed concurrently.

The challenge of synthesis

  • The evidence base is vast — many factors, many studies, many interactions

  • Hard for anyone to judge which driver is most important in a particular context

  • Requires analysis that combines drivers simultaneously so relative influences can be assessed

This is exactly what was missing in Brighton and Hove. A single strand of evidence (segregation) was treated as the whole story, to the exclusion of other — potentially more impactful — factors.

Brighton and Hove: Context

The city’s secondary school landscape

Figure 1

10 secondary schools — a unique admissions system

  • 6 community schools (LEA-controlled), 2 academies (BACA, PACA), 2 church schools

  • Catchment areas with lottery tie-break for oversubscribed schools (since 2008)

    • Replaced distance-based allocation
    • Very unusual in England

  • In 2023, FSM priority added — already affecting mixing without opposition

  • Brighton & Hove is already in the top third nationally for integration of disadvantaged/non-disadvantaged pupils

The 2024 consultation: the proposals

What was proposed:

  • Reduce PANs at popular central schools (Blatchington Mill, Dorothy Stringer)
  • Reserve 20% of places for out-of-catchment children
  • Redraw catchment boundaries in the east

Council’s stated objectives (Dec 2024 Cabinet papers):

“reducing some schools’ barriers to success” for disadvantaged pupils

“a more mixed pupil intake creates better outcomes for disadvantaged pupils”

Stated premise: results in the city were “driven by economic advantage” — so redistribute → narrow the gap

The 2024 consultation: the response

Public reaction:

  • “Strong preference for improving existing schools rather than redistributing students”
  • Concerns about community cohesion and student wellbeing
  • National media coverage of vocal objections

The institutional context:

  • Labour majority + Leader & Cabinet system
  • Compressed consultation timeline
  • Limited scope for deliberative engagement
  • Privileged access for some stakeholder groups over others in “policy formation”

The key questions this raises

  1. Is the premise correct? Is attainment in Brighton & Hove really “driven by economic advantage”?

  2. Is this the right lever? Even if concentrations of disadvantage matter, are they the most important factor?

  3. How is the city actually performing once structural factors are accounted for?

  4. What are the risks of pulling this particular lever — including second-order effects?

We can answer all of these with DfE open data.

Data & Methods

DfE open data

Data sources:

  • School performance (Attainment 8, Progress 8)
  • Pupil absence and attendance
  • Pupil characteristics (FSM, EAL, prior attainment)
  • School workforce (retention, sickness, pay)
  • Ofsted ratings
  • Admissions policies
  • School characteristics

Our panel:

  • 13,419 school-year observations
  • 3,523 academies and maintained schools
  • 152 Local Education Authorities
  • 4 academic years (2021–22 to 2024–25)
  • Post-COVID, post-imputed grades

All publicly available. All linkable. All underutilised.

What is the ‘model’?

  • Statistical explanation of how GCSE Attainment varies between schools after accounting for things we know affect attainment
  • Raw league tables compare finishing positions, our model adjusts for the wind

Non-linear effects: log-transformations and importance of local context

Figure 2

The same percentage-point change produces very different effects depending on where a school starts. Where you sit on the curve matters enormously for policy.

Model specification

Multilevel linear mixed effects model — schools nested within LEAs within regions:

\[ \log(\text{ATT8}_{ij}) = \beta_0 + \sum_{k=1}^{9} \beta_k \, x_{kij} + u_{\text{year}} + u_{\text{Ofsted}} + u_{\text{region}} + u_{\text{LA|region}} + \varepsilon_{ij} \]

Fixed effects (predictors):

  • log(% FSM-eligible)
  • log(% Overall absence)
  • log(% English as an Additional Language - EAL)
  • % Low prior attainment (KS2)
  • Selective admissions (dummy)
  • Gorard Segregation Index (LA-level)
  • Teacher retention rate
  • Leadership pay proportion
  • log(Teacher sickness days)

Random effects (grouping):

  • Academic year
  • Ofsted rating (4-category)
  • Government Office Region
  • LEA nested within region

Three models:

  • All pupils
  • Disadvantaged pupils
  • Non-disadvantaged pupils

National Results

How good are the models?

Model Marginal R² Conditional R²
All pupils 0.63 0.77
Disadvantaged 0.56 0.73
Non-disadvantaged 0.62 0.79

\(R^2\) measures how much variation in Attainment 8 is explained by the model

Can be interpreted as a % - e.g. 0.8 = 80%

School-level attainment is remarkably predictable from a small number of structural factors.

~80% of variation explained for all pupils; ~70–80% for subgroups.

  • Disadvantaged pupils: more variation from unmeasured factors (individual resilience, specific interventions)
  • Non-disadvantaged: tighter, more predictable

The mediation story: building the model step by step

Table 1
M1: FSM only M2: + Absence M3: + Prior att.
FSM coefficient -0.201 -0.122 -0.073
Absence coefficient — -0.364 -0.287
Prior att. coefficient — — -0.006
R² 0.393 0.612 0.669

Mediating variables sit on the causal pathway, not including can inflate apparent relationships: FSM → Absence → Attainment

Adding absence halves the FSM coefficient. Adding prior attainment reduces both further. With just 3 variables → two-thirds of variation explained.

What about segregation?

  • The Gorard Segregation Index is not statistically significant in the full model

  • Once a school’s own FSM levels, attendance, prior attainment and geographic location are accounted for → no additional predictive power

  • Any harm segregation causes is fully mediated by the other variables

  • This doesn’t mean segregation is harmless — it means its effects operate through absence, prior attainment, and other measured factors

  • We include it because it formed much of the justification for the Brighton & Hove proposals

Relative variable importance

Figure 3

Key finding: absence dominates

Absence is the most powerful predictor of school-level attainment by a considerable margin — its coefficient is roughly 2.6 times larger than concentrations of disadvantage.

  • For disadvantaged pupils, nearly half of school-level performance variation is explained by attendance alone

  • Absence affects disadvantaged pupils far more than non-disadvantaged

  • Disadvantaged pupils lack safety nets (tutors, engaged parents, revision guides) — they are more reliant on classroom instruction

  • Attendance is the ultimate equity lever

  • On the council’s own framing — “reducing some schools’ barriers to success” for disadvantaged pupils — attendance is the largest, best-evidenced barrier the model can quantify

The contentious finding: concentrations of disadvantage (1)

For non-disadvantaged pupils:

  • Higher FSM → lower attainment

For disadvantaged pupils:

  • Higher FSM → slightly higher attainment (opposite BHCC consultation statement)
  • After controlling for absence & prior attainment
  • Robust across years and alternative specifications - echos 2015 DfE evidence

Why? Schools with higher concentrations likely develop specialised support systems — targeted use of Pupil Premium, vocational pathways, specialist staff.

Effect shrinks in the full multilevel model → operates through LEA/Ofsted factors, not concentration per se.

Impact Even before negative impacts of redistribution policy (longer journeys etc.), BHCC policy likely to negatively impact disadvantaged attainment at city level.

The contentious finding: concentrations of disadvantage (2)

  • To be clear: We do not advocate for active concentration of disadvantaged pupils.

  • But deconcentration policies premised on the assumption disadvantaged students inevitably fare worse in higher-disadvantage schools are not supported by this evidence.

  • Where deconcentration policies create other negative externalities, classic example of pulling the wrong lever

Brighton & Hove: The Evidence

How is Brighton & Hove actually performing?

Figure 4

Brighton & Hove: a story of success, not failure

LEA rankings (after structural adjustment):

Pupil group Rank / 152
Disadvantaged 7th
Non-disadvantaged 5th
All pupils 4th

The positive random intercept ≈ +2 GCSE points above what structural factors predict.

This was entirely unknown at the time of the 2024 consultation and completely absent from the public narrative.

The starting point was a narrative of failure. The evidence says the opposite: this is one of the highest-performing LEAs in England.

Rather than “what is the city doing wrong?”, the question should be “what is it doing right?”

But there is a problem — and it isn’t segregation

FSM/Disadvantaged Mixing (the lever chosen):

  • City mean: 28.8%
  • National mean: 28.3%
  • National percentile: 52th
  • Completely unremarkable

Absence (the lever totally ignored):

  • City mean: 10.8%
  • National mean: 8.2%
  • National percentile: 99th
  • Among the very worst in England

Brighton & Hove has near-average disadvantage but near-worst absence. The city’s strong underlying performance is being dragged down by an absence problem at the extreme of the national distribution.

Absence league table

Table 2
Local Authority 2024-25 2022-23 2021-22 2023-24
Knowsley 10.9% (152/152) 11.8% (149/152) 12.6% (152/152) 11.9% (151/152)
Brighton and Hove 10.8% (151/152) 10.5% (137/152) 10.8% (144/152) 11.0% (143/152)
Newcastle upon Tyne 10.6% (150/152) 12.9% (152/152) 12.2% (151/152) 11.4% (150/152)
Southampton 10.5% (149/152) 10.6% (139/152) 10.1% (119/152) 11.0% (145/152)
Bradford 10.3% (148/152) 11.9% (150/152) 11.5% (149/152) 11.2% (147/152)
Plymouth 10.2% (147/152) 10.8% (144/152) 11.1% (147/152) 11.3% (149/152)
Middlesbrough 10.1% (146/152) 12.9% (151/152) 11.8% (150/152) 12.2% (152/152)
Sefton 10.0% (145/152) 10.3% (129/152) 10.1% (122/152) 10.5% (133/152)
Devon 9.9% (144/152) 10.9% (145/152) 10.8% (145/152) 10.8% (139/152)
Dorset 9.9% (143/152) 10.2% (127/152) 10.3% (134/152) 10.7% (137/152)
Halton 9.8% (142/152) 10.3% (131/152) 10.1% (118/152) 11.0% (144/152)
Blackpool 9.8% (141/152) 11.1% (147/152) 9.8% (103/152) 11.3% (148/152)
Hartlepool 9.8% (140/152) 11.1% (148/152) 10.2% (130/152) 10.5% (134/152)
Bristol, City of 9.6% (135/152) 11.0% (146/152) 10.6% (141/152) 11.0% (142/152)
Gateshead 9.5% (130/152) 10.8% (143/152) 11.4% (148/152) 11.2% (146/152)
St. Helens 9.1% (121/152) 9.6% (103/152) 10.7% (143/152) 10.0% (122/152)
Torbay 8.2% (75/152) 10.2% (125/152) 10.9% (146/152) 10.3% (126/152)

Accelerating returns: comparing the two levers

Figure 5

What would closing the absence gap deliver?

Table 3
School Current Absence % National Avg (Target) Reduction (pp) ATT8 gain (all) ATT8 gain (disadv.)
Longhill High School 16.7 8.2 8.4 5.3 6.1
Brighton Aldridge Community Academy 14.4 8.2 6.2 4.6 6.2
Hove Park School and Sixth Form Centre 12.9 8.2 4.6 4.1 4.7
Portslade Aldridge Community Academy 11.3 8.2 3.1 3.0 3.2
Blatchington Mill School 10.8 8.2 2.6 3.0 3.3
Dorothy Stringer School 9.8 8.2 1.5 2.0 1.9
Varndean School 9.0 8.2 0.8 1.0 1.0
Cardinal Newman Catholic School 8.5 8.2 0.3 0.3 0.4

Bringing absence to the national average → predicted gains of 3–5 GCSE points at the worst-affected schools. FSM reductions → closer to 1–2 points.

Observed vs predicted: disadvantaged pupils nationally (2024–25)

Figure 6

Alternative league tables: value-added

Disadvantaged Pupils: Value-Added Rankings (ATT8 points vs prediction)
Rank School 2021-22 2022-23 2023-24 2024-25 Mean
1 Brighton Aldridge Community Academy +3.2 -1.3 +4.1 +5.2 +2.8
2 Varndean School +3.2 +3.6 +4.5 -1.9 +2.3
3 Dorothy Stringer School +2.9 +3.2 +0.3 +1.7 +2.0
4 Cardinal Newman Catholic School +2.4 -5.0 +6.5 +2.5 +1.6
5 King's School +5.8 +0.9 +4.4 -5.4 +1.5
6 Blatchington Mill School -5.4 +1.3 +5.2 +3.4 +1.1
7 Hove Park School and Sixth Form Centre +2.4 -1.7 +1.3 +0.5 +0.6
8 Portslade Aldridge Community Academy +2.6 -4.0 +4.7 -1.3 +0.5
9 Patcham High School +3.8 +1.6 -11.9 -3.7 -2.6
10 Longhill High School -4.6 -6.2 +0.8 -1.4 -2.9

The BACA paradox

What people thought:

  • ‘Requires Improvement’ Ofsted
  • Low raw attainment scores
  • → narrative of “Avoid this school”
  • Equity in Education campaign: parents opt for Patcham instead

What the data shows:

  • Best performing school in the city for disadvantaged pupils
  • +5 GCSE points above national peers (2024–25)
  • April 2025: Ofsted upgraded to ‘Good’ in all areas

Patcham — “the safer choice”:

  • Rated ‘Good’ by Ofsted
  • Comfortable mid-table in raw terms
  • But negative value-added for disadvantaged pupils
  • 4-year average: –2.6 GCSE points

Parents, acting rationally on available information, may have been choosing the school with lower value-added for their children — if that is the most important factor. They were operating in a knowledge vacuum.

Attendance as a school-level lever

  • Patcham is one of only two schools in the city with attendance better than the national average

  • This is an actionable lever at the school level

  • BACA exemplifies the national finding: schools with higher disadvantage can develop specialised, effective support

  • Every school in the city could learn from Patcham on attendance — and from BACA on supporting disadvantaged pupils

The city needs to disaggregate the attendance problem — different causes require different strategies: family holidays, SEND challenges, transport barriers, caring responsibilities.

Absence isn’t one thing — it’s two

  • Absence in our headline model is a single number, but conceptually it sits in two roles at once

  • Part of it is structural — driven by intake, family circumstances, area health, deprivation, SEN, EAL…

  • Part of it is school-managed — pastoral systems, attendance officers, parental engagement, ethos around getting children into class

  • A two-stage decomposition1 separates these:

    • Stage 1: model expected absence from intake variables only
    • Stage 2: refit attainment using expected absence; what the school adds (including its attendance management) flows into the value-added residual

Most of school-level absence variance nationally is structural. The school-controllable share is the smaller part — but it is the share local policy can actually move.

Where do schools sit on the joint-signal plane?

Figure 7

What this view changes

  • BACA sits firmly in Q2 — adding value despite worse-than-predicted attendance. Closing its residual-absence gap would compound an already-strong pedagogical signal.

  • Patcham sits in Q4 — underperforming despite better-than-predicted attendance. The simple “fix attendance” story isn’t available here.

  • Schools in Q3 (underperforming + attendance worse) are the cleanest single-lever targets — both signals point the same way.

  • Schools in Q1 (adding value + attendance helping) are where the city should look for practice worth sharing.

Reading value-added alongside residual absence separates “this school adds points to attainment” from “this school’s pupils attend more than their intake predicts” — two distinct signals that single-residual league tables fold together.

Discussion & Reflections

The danger of single-lever policy thinking

  • The council’s proposals were built on the premise that attainment was “driven by economic advantage”

  • This premise is false in the context of the DfE’s own research (Macleod et al. 2015) — our modelling confirms this was at best a flawed understanding

  • Concentrations of disadvantage are one of the weaker levers available

  • For disadvantaged pupils specifically, the direction of effect runs contrary to the policy assumptions - across England over 4-years of data, disadvantaged pupils get better results in schools with higher concentrations of disadvantaged pupils

  • The policy was pulling a lever with relatively little mechanical advantage while ignoring one with considerably more

Second-order effects: the risks of the chosen lever

  • The chosen policy if implemented as intended requires longer journeys for many children

  • Thomson (2023): pupils who travel further to school are absent more often

  • Given absence is the city’s most acute problem and the strongest predictor of attainment…

A policy that even marginally increases absence could be counterproductive in attainment terms. The claim of “almost no cost” for social mixing overlooks these second-order effects.

Following opposition and appeals: PAN reductions rejected, 20% out-of-catchment → 5%. But the policy reverberations continue through parental choices already made. Little displacement on 2026 offer day luck not judgement!

What we don’t know

The LEA random effect is large and significant — something about Brighton & Hove’s schools is delivering results well above prediction. But what?

  • High levels of parental engagement and aspiration? (well-educated population)
  • Quality of school leadership and governance?
  • Effective local authority support services?
  • Collaborative relationships between schools?
  • Community factors difficult to quantify?

Identifying sources of over-performance is a key research question. But policy interventions that risk disrupting a well-functioning system without understanding what makes it function well carry substantial downside risk.

What should Brighton and Hove do?

The council has recently agreed to a cross-party working group on schools — a positive step. Our strongest recommendation:

Priority one: a deep dive into absence. The city has the 2nd worst absence rate in England but near-average disadvantage. This is not inevitable — it is, in principle, solvable.

The absence problem needs disaggregating — very different situations exist within the headline statistics:

  • A few days missed for a family holiday by an otherwise well-attending pupil
  • A SEND student who struggles with the school environment but has strong family support
  • A student with caring responsibilities living far from school, struggling with bus journeys
  • Post-pandemic disengagement and mental health challenges

Each requires a different strategy. Schools like Patcham — one of only two with attendance below the national average — offer lessons in what works locally.

Lessons for other LEAs

  • Every LEA has a different profile of challenges — different positions along non-linear curves

  • What works in one context may be irrelevant or counterproductive in another

  • The DfE open data provides the raw materials for local understanding

  • But there’s a crucial gap between raw data and actionable intelligence

  • National ambitions will founder if pursued locally without adequate analytical infrastructure

The Policy Simulator Tool

Bridging the gap

What it does:

  • Search for any school in England
  • See key statistics and contextual benchmarks
  • Compare observed vs predicted attainment
  • Simulate the impact of changes to absence, FSM, workforce etc.
  • Explore LA-level patterns and typologies
  • Find a school’s contextual “twin”

Built with:

  • R Shiny
  • Same multilevel models as this analysis
  • DfE open data (updated annually)
  • Development accelerated by Claude (Anthropic) — AI4CI funded

Not the finished article — but a demonstration of what’s possible when open data + appropriate methods + accessible tools come together.

The simulator in action

School Finder → select any school → view contextual statistics → send to the Policy Simulator

adam-dennett.shinyapps.io/School_Attainment_Policy_Simulator

Adjusting policy levers

Non-linear effects mean the same adjustment produces different impacts depending on the school’s starting position

Conclusions & Recommendations

Key takeaways

  1. School-level attainment is remarkably predictable (~80% variance explained) from a handful of open data variables

  2. Absence is the dominant predictor — 2.6x more important than concentrations of disadvantage

  3. For disadvantaged pupils, the concentration effect runs in the opposite direction to common assumption

  4. The Gorard Segregation Index is not significant once other factors are controlled for

  5. Brighton & Hove is one of the highest-performing LEAs — but has the 2nd worst absence in England

  6. The 2024 policy was pulling the wrong lever — targeting the weaker factor while ignoring the stronger one

  7. Crude metrics (raw league tables, single-word Ofsted) mislead parents and policy makers

Recommendations

For the DfE:

  • Invest in analytical infrastructure at the local level
  • Fund LA analytical capacity or develop nationally maintained benchmarking tools
  • Go beyond raw league tables

For LEAs:

  • Resist single causal narratives
  • The factors are multiple, interacting, non-linear, context-dependent
  • Create institutional space for genuine deliberation before irreversible decisions
  • Prioritise attendance as a policy lever

For schools, governors & parents:

  • Contextualised benchmarking offers a fairer basis than raw scores
  • Demand better information before making choices
  • School ‘quality’ ≠ raw attainment

The final word

The caution: Policy made in haste, anchored to a single causal narrative and implemented without adequate regard for evidence, risks doing more harm than good — particularly when it disrupts a system that is working well.

The encouragement: The raw materials for better policy already exist in DfE open data. With the right analytical tools and a commitment to evidence over ideology, it is possible to understand with considerable precision what drives attainment in any local context — and where the most productive interventions lie.

The challenge: Ensuring this understanding reaches decision makers in a usable form before decisions are made — not after.

CASA logo
AI4CI logo