Engineidle · awaiting next release
Dashboard/Methodology

Methodology

Every formula, every dataset role, every limitation — auditable in one place.

Datasets and their role

DatasetRole
notineducationemploymentortrainingjm26.xlsxLatest NEET signal
notineducationemploymentortrainingod25.xlsxPrevious release comparison
neetcitableupdatedjm26.xlsxSampling variability / confidence
analytical_annexe_data_tables.odsExplanatory signal library

NEET Pressure Index — weighted composite

All sub-scores clamped 0–100; weights sum to 1.0. Current value: 70 / 100 (elevated).

  • Level (NEET rate) Higher rate means more participation pressure.w 30% · score 68
  • Direction (year-on-year) Rising YoY adds pressure.w 25% · score 98
  • Persistence (consecutive QoQ) Sustained rises are stronger signals.w 15% · score 50
  • Composition (inactive share) Higher inactivity = deeper detachment risk.w 20% · score 68
  • Signal confidence Movement relative to sampling variability.w 10% · score 39

Confidence calculation

movementVsCi = |QoQ change in people| / 95% sampling-variability CI (people)

  • · < 0.5× → cautious (likely noise)
  • · 0.5–1× → emerging (below CI but non-trivial, monitor)
  • · 1–2× → moderate (signal exceeds CI)
  • · ≥ 2× → strong (more than twice CI)

Current movement 54,769.625 people vs CI ±70,000 people → 0.78× → emerging.

Signal triggering rules

  • · Participation Pressure: always evaluated. Severity by NEET rate band.
  • · Inactivity-Driven Pressure: inactive share ≥ 55%.
  • · Labour-Market Absorption Risk: unemployed share > 40% AND YoY > +1%.
  • · First-Rung Access Risk: rate(18–24) − rate(16–17) > 8 pp.
  • · Hidden NEET Visibility Gap: headline rate > 11%.
  • · Long-Term Detachment Risk: ≥ 2 consecutive QoQ rises AND inactive share ≥ 50%.
  • · Easing: YoY < −2% AND QoQ < 0.

AI / RAG briefing logic

The AI layer receives a structured evidence payload (see AI Briefing page) and converts it to readable narrative. It MUST cite only values in the payload; it MUST NOT invent causes. Today this is a deterministic template; the in-house model replaces only the language step — interface remains identical.

What this system cannot do

  • · Predict whether any individual will become NEET.
  • · Determine benefits eligibility or take individual-level decisions.
  • · Replace local-authority diagnosis or human policy judgement.
  • · Operate without human review of every briefing before circulation.

Why human review is required

Signal rules and the confidence layer are transparent but limited. Reviewers correct for context the aggregate data cannot see (local programme starts, policy changes, sector shocks). Every briefing is DRAFT until a named reviewer approves it for export.

Sources displayed across the system: ONS LFS release JM26 (January to March 2026); previous OD25.