ONS Education & Childcare statistics collection

It is the Office for National Statistics' landing ground for everything it publishes on early years childcare, school and college education, higher education and adult learning - qualifications, personnel, safety and well-being - organised as Spotlight features, a datasets list and a publications list of bulletins and articles.

The collection's centre of gravity is breadth of measurement rather than depth of history in any single series. On one shelf: Census 2021 qualification products, including local-characteristics tables for Output Areas, detailed-characteristics tables for merged wards and Middle layer Super Output Areas, workforce qualification levels, and qualifications by country of birth. On another: Growing Up in England, a linked administrative series split into personal characteristics, household characteristics and educational attainment, vulnerable groups, and Individualised Learner Record extracts.

Then the analytical layer: Longitudinal Educational Outcomes studies that follow GCSE cohorts from 2012-13 to age 22 and cut attainment and earnings by town; the COVID Social Mobility and Opportunities study's Wave 1; childcare accessibility computed from LSOA up to regional level; public-service education productivity indices; and reference tables quantifying the free school meals earnings gap.

Around the datasets sit the bulletins that get quoted: Young people not in education, employment or training (NEET), UK - November 2024 estimated 946,000 people aged 16-24, with a May 2026 edition published as official statistics in development - and Education, England and Wales: Census 2021. Articles cover free-school-meal recipients' earnings, SEND experiences, remote schooling during the pandemic, and the education background of looked-after children.

Aggregate products resolve to one honest shape: a geography code, a period label, a measure name where needed, and a value. A childcare row is an LSOA code, the year 2023, the measure 'places per child aged 7 and under', and a number like 0.87. A Census qualification row swaps the measure for a highest-qualification category and the value for a count. That uniformity is deliberate - ONS area codes appear unchanged whether the geography is a ten-household Output Area or the whole of England - which makes stitching products together a join on keys you already have rather than a mapping exercise.

The microdata products break that pattern on purpose. Growing Up in England links individual pupil records to household characteristics and attainment; COSMO Wave 1 surveys a cohort born into the pandemic; the Census teaching file ships individual-level records built for classroom use. These carry person-level constructs, not area-code-plus-value, and their schemas differ by extract.

Two disciplines travel with every product. First, observation_status: ONS marks suppressed, low-reliability and withheld values explicitly, so gaps are labelled rather than silent. Second, release metadata - each product page stamps a release date and a next-release notice, which matters because cadence varies enormously across the collection, from static Census cross-sections to twice-yearly NEET editions.

Sample rows above show product-level attributes as observed in the collection index this session; row-level values arrive with your sample.

The dictionary above covers the spine shared by most aggregate tables: area_code and area_name locate the row, date fixes the period, variable names the measure in long-format layouts, value carries the number, observation_status explains anything missing or caveated, and release_date ties the record to its publication.

Confidence here is honest rather than absolute: these definitions are inferred from ONS standard table layouts rather than read off a single published data dictionary, because the collection has no unified schema document - each product documents itself. Column headers inside individual workbooks occasionally vary from the canonical names above, particularly in older series and bulletin chart-data extracts.

The person-level products sit outside this dictionary by design. GUiE extracts, COSMO Wave 1 instruments and the Census teaching file use their own variable sets, documented per product; ask and we include the relevant schema with your sample so nothing surprises you at mapping time.

Geographically this is a UK collection with English weight. National bulletins cover the United Kingdom; most datasets stop at England or England and Wales, reflecting where education policy is devolved. The ladder runs from Output Areas - the finest Census geography - through Lower and Middle layer Super Output Areas and merged wards, up to local authorities, regions and country. If your analysis needs Scotland or Northern Ireland education detail, this collection will not carry it beyond UK-level aggregates; treat that boundary as real, not incidental.

Temporally the products refuse to synchronise, and that is informative rather than annoying. Census 2021 tables are fixed cross-sections. LEO analyses anchor to GCSE cohorts - 2012-13 onwards - and mature as each cohort ages to 22. Childcare accessibility is referenced to 2023 with its next release marked to be announced. NEET publishes at least twice yearly, with the newest editions carrying the 'official statistics in development' flag while methodology settles. Productivity indices run as continuous annual series.

Grain follows purpose. Where the question is individual lives - vulnerable children, pandemic cohorts - the collection offers genuine microdata under controlled terms. Where the question is system state, it offers tidy aggregates. Both live in one catalogue, which is precisely why teams that need either end of the granularity spectrum end up pulling from it.