Roadmap

What v2.1 does now, and what comes next.

Done in the first pass

• Value Added v2.0 specified on the review’s recommendations and fitted for all, disadvantaged, non-disadvantaged pupils, girls and boys
• Value added against England and within the authority, shrunk, with 95% intervals
• Every school in England, explored by region, combined or mayoral authority, or local authority, on a map with a sortable table, and a downloadable file
• Pages for Sussex, its three authorities and twelve districts, with the map, the girls/boys split, a v2.0 decomposition and the latest Ofsted judgement
• Comparison with the published v1 measure throughout
• Absence above or below what the intake predicts, on the v2 intake terms, on every map and in every profile
• The original Sussex documents’ analyses refitted on v2.0: the decomposition, the district test, trusts, the joint-signal plane, absence scenarios, comparator profiles, league tables and questions for discussion; a district summary page
• Value added over time: every school’s yearly v2.0 value added, trajectories in the explorer and on every area page, and a centred moving average over one, three or five years around a chosen year

Done since: absence, schools and twins

• What goes with school absence. A page that starts from a map of the UK research (with links to the sources), shows the trend by reason since 2016/17, and models seven kinds of absence in 2025/26 on every secondary school’s intake, needs and setting, with each school nested in its authority and region. Shapley shares say how much each group of characteristics explains, and the full models (coefficients, random effects, fit) are on the page.
• The school’s part. What the models attribute to “the school” is mostly selection through admissions; what is left between similar schools is wide and lasts from one year to the next. A message for heads and governors, nationally and on every area page.
• Absence on every Sussex page. Each authority’s and district’s absence against England, its neighbours and the tier above, what in its intake raises or lowers it, and its schools against similar schools in both years.
• Find a twin. Choose a school and find the schools in England most like it in intake and setting, or the look-alikes that do better on absence, attainment or value added, on a chart and a map, with fact cards side by side. Linked from the absence page and every area page, and every comparison has its own link.
• How school absence changes (started). Every secondary school followed from 2016/17 to 2025/26 through academy conversion and changes of trust: what is lasting and what moves, whether schools keep their place, who rose most since the pandemic, whether absence moves when a school’s own intake does, and event studies of becoming an academy and of a change of trust.
• The explorer. Ofsted colours that work across both inspection regimes, in each palette, and the spread of schools on the map beneath it, as on the Sussex pages.

Next

• Policy simulator v2.0. The Pull the Right Lever simulator rebuilt on the v2.1 models, in the browser: absence and the intake terms as levers (with admissions and sixth form for one school), all five groups at once with the disadvantage gap, each lever’s contribution, and an area mode that applies a change to every school in a district, authority, region or England.
• Adding a new year. When DfE publishes a new year’s KS4 results, add it to YEAR_CONFIG in pipeline/helpers.R, run R/00_fetch_data.R, then the build steps. The v2.1 model takes its years from the data, and no longer needs the workforce file, which was the slowest to arrive. See the README.
• A self-contained data pipeline for the model. R/00_fetch_data.R builds every model input from public sources (DfE, GIAS, ONS, nomis, postcodes.io, Ofsted), with the scripts in pipeline/. Rebuilt from the same downloads it reproduces the inputs, and every value-added figure, exactly.
• The Sussex area analyses, self-contained too. The area pages still reuse the original project’s area analyses (a snapshot in data/area/), its Sussex governance and its absence reasons. Port its R/15, R/17 and R/19 next.
• Update 2025/26 with the revised KS4 release (February 2027). 2025/26 is shown now, provisional, predicted from the fitted model on DfE’s October results with absence bridged and the unpublished intake data carried forward (R/05_provisional_2526.R). Rerun it on the revised results; then, once the year’s census, SEN and mobility data are out, fold 2025/26 into the model as a fifth year.
• v2.1 candidates tested across every group: the illness share of absence and an 11–16 flag (the methodology), with a chart of what matters most in the model.
• v2.1 published. The 11–16 flag is in the expectation (a school’s structure, like its admissions policy). Value added allowing for the kind of absence is published beside it, not in it, because what is authorised as illness is partly a school’s own coding.
• Neighbourhood types (OAC). The 2021/2 Output Area Classification profiled over each school’s 150 nearest Output Areas at all three tiers. It reproduces neighbourhood education, adds nothing to results or to value added within authorities, and describes the places behind the authority effects for disadvantaged pupils (multicultural urban areas above, small-town, coastal and former industrial areas below). Groups are enough; subgroups add little.
• Geographical effects across England. How much of value added is region and local authority, for every group; a map of the authority effects; region averages; and what goes with them.
• Neighbourhood covariates (experiment). Ten Census 2021 measures and each school’s own White British share added to v2.1. For disadvantaged pupils, the ethnic pattern is mostly composition (pupils, not place); neighbourhood deprivation beyond qualifications is place. Together they take about a fifth off the authority effects and over a third off the regional ones for disadvantaged pupils; about a third of Brighton and Hove’s effect.
• v2.2 candidates tested across every group and year. Neighbourhood household deprivation helps every group in every year (recommended for v2.2). Pupils’ ethnicity (school White British share, each January) matters almost only for disadvantaged pupils, where it is large; recommended beside the headline, not in it.
• Value Added v2.2 (agreed, for a future release). The approach from the v2.2 candidates is agreed but not yet built: refit every group with neighbourhood household deprivation in the expectation beside neighbourhood education; publish value added allowing for pupils’ ethnicity beside the headline, labelled, not in it; and before release, test detailed ethnic groups rather than the White British share alone. Until then the site publishes v2.1.
• Parents, not adults. Neighbourhood education rebuilt from custom Census tables as graduates among parents (adults 25-64 with dependent children, MSOA). All-adult graduate shares hide a 12-point gap in parents between deprived and comfortable graduate-rich neighbourhoods, and the young-adult penalty almost disappears. Parents alone is no better a term; parents and other adults together help every group a little. Deprivation survives.
• v2.2 candidate: split neighbourhood education into parents and other adults, and follow up why disadvantaged pupils do better near graduates without children.
• What causes school-level absence, continued. The cross-section and change over time are under way (above). Next: each school’s own path relative to England on the area pages and in Find a twin, whether high-illness places such as Brighton and Hove record absence differently, the severe tail on its own, whether particular trusts’ schools move together, and absence against intake by kind as a candidate for a future version of the model.
• Twins, further. The matching is weighted by what matters for the measure chosen (equal weights still available), the fact cards show value added and absence year by year against the twin and England, and a comparison prints cleanly for a governors’ meeting.
• Distance to school: the straight-line check. The distance from every neighbourhood in England to its nearest state secondary, by deprivation, in every district and authority. Across England deprived neighbourhoods are usually nearer; in Brighton and Hove they are about a third further away. Absence goes with distance by authority, more for disadvantaged pupils, and by school, mostly in illness and persistent absence. It builds on FFT Education Datalab’s finding that pupils who live further away are absent more often.
• How firm is the link with absence? Six years, eleven measures of distance and thousands of models for all three groups of pupils, a falsification test on primary absence, DfE’s absence by home neighbourhood within each authority, and changes in the school network. The link is consistent and stronger for disadvantaged pupils, but most of its size between authorities is ethnic mix; a smaller link remains in towns and cities, within authorities and between schools. Every source and calculation is in the methodology.
• Journey times on the real networks. Walking and bus times with r5r from every Output Area in Brighton and Hove and eight comparator places, with changes of bus, the penalty for missing a journey, school buses and the climb each way. Brighton and Hove stands out on real journeys as it does in straight lines. None of these shows a clear general link with absence yet; Longhill stands out on missing the bus.
• Journey times for the mostly urban authorities. Routing enough places to rerun the absence models on time rather than distance, and so test whether time says more than straight lines.
• Points of interest around bus changes and along walking routes. From OpenStreetMap, within a couple of hundred metres of change points and along routes, to test directly whether a longer urban journey matters because it passes more places to stop. More general than shops, which may not sell to children in school hours. The routes and change points are ready for it.
• School bells. Published start times for each school, in place of the time inferred from the last school bus, for the penalty for missing one.
• Distance to school: how far each school’s pupils travel. A better measure than each school’s nearest children, such as a gravity model of who goes where, built from published sources.
• Distance to school: does it matter more in towns and cities? The hypothesis: distance matters for attendance more than is understood, and may matter more in towns and cities, where a longer journey passes more places to stop on the way (intervening opportunities), than in rural areas, where long journeys are the norm and often by school transport. Between authorities the link is strong in mostly urban places and absent in less urban ones, but within authorities and by school it is similar in both, so it needs journey times and data on school transport.
• Independent schools. What is published for them, term by term against v2.1; why their Attainment 8 cannot be compared (IGCSEs score nothing, and KS4 and A level results are unrelated); a test showing a reduced specification gets value added badly wrong even in the state sector; and the comparison that does exist, DfE’s A level value added. Verdict: no value added at 16 is possible from published data.
• Independent schools: IGCSEs. DfE’s 16-19 guidance checked: A level value added counts approved GCSEs only, so IGCSEs are left out of the baseline (most independent students still have some GCSEs). Attainment 8 completed from its published parts: the average grade in counting GCSEs, validated on state schools (R² 0.88-0.90) and against independent schools’ own A level results (r 0.84, against 0.03).
• Independent schools, next. A level value added over three years and by subject; the completed Attainment 8 over several years; and the two sectors’ sixth forms side by side by area.
• Areas beyond Sussex. The area template works for any authority and its districts once boundaries are added.
• Monthly Ofsted refresh. A GitHub Actions workflow picks up Ofsted’s new release on the 20th of each month, updates every school’s latest judgement without refitting, rerenders and commits.
• Publish v2.1 alongside v1 in the original reports, labelled, with a note on how rankings change.