Eurostat Education & Training Database (educ_* tables)

One theme, six shelves, all speaking the same classification language. The participation shelf counts heads: enrolment by ISCED 2011 level, sex and age, including the flagship EDUC_UOE_ENRA02 cube with 754,470 observations running 2012 through 2024, and finer cuts such as EDUC_UOE_ENRA20, which asks what share of three-year-olds sit in a classroom - Germany's answer for 2023 was 91.7 percent.

The people shelf measures the workforce: teaching staff counts and characteristics. The outcomes shelf follows them out the door - graduation rates, attainment levels, and early leavers from education and training, the indicator every EU employment strategy quotes. Adult learning tracks participation by people already in the labour market, and the languages shelf records which foreign tongues schoolchildren study. Finally the money shelf: public expenditure on education in absolute terms, per student, and relative to GDP.

About 112 of the roughly 156 EDUC-prefixed dataflows sit in the UOE (UNESCO-OECD-Eurostat) joint collection, meaning OECD members and UNESCO institutes supply their figures into the same harmonised frame. Data arrive from national statistical institutes under agreed definitions, so a percentage point in Lisbon weighs the same as one in Helsinki.

A row is a value plus its complete address. Take the EU27 aggregate for 2022: 87,316,666 people enrolled, carrying geo=EU27_2020, time=2022, unit=NR. Split it once and you get 43,614,131 males under isced11=TOTAL; split it again and there are 7,308,710 women aged Y20-24. Same cube, same keys, different slices - which means a query pattern written once keeps working as you add dimensions.

The tab-separated export shows the other face of the same data: one line reading A,NR,TOTAL,F,ED0,AT followed by a value per reference year - Austria's early-childhood enrolment for girls across successive years laid out left to right. Dimensions lead, values trail, colons separate flags. Whichever shape you prefer, the observation status travels with the number: d flags a country whose definition differs from the standard, b marks a series break. Those two characters are why you can chart twelve years of Austrian pre-primary counts and know exactly which kink is policy and which is a redefinition.

One structural note worth planning around: most tables stop at the national border. If your analysis needs regions, provinces or individual schools, pair this database with something sub-national rather than expecting it here.

Nine fields cover the core cubes: freq, unit, age, sex, isced11, geo, time, value and status. Six of them are dimensions you filter on, one is the measure you read, and the last tells you how much to trust it. The dictionary above pins each field to a type, a definition and a real example drawn from live records - ED5 for short-cycle tertiary on the ISCED ladder, EU27_2020 for the union aggregate, d for a definition drift flag.

Two conventions repay attention. First, units travel per-row rather than per-table, so the same query can return NR counts and PC shares side by side if you ask it to; filter on unit before you chart. Second, age codes are strings with grammar - Y10-14 style bands, Y_GE65 for open-ended upper bounds, UNK for unknown - so sorting them alphabetically produces nonsense and parsing the prefix produces sense.

Geographically: EU27, EFTA, EU candidate countries and the United Kingdom, with aggregates such as EU27_2020 and the euro area pre-computed so you can quote union totals without summing member states and arguing about rounding. Temporally, depth varies by table by design - EDUC_UOE_ENRA02 spans 2012 to 2024, expenditure series reach back further, and the latest reference year typically lands two years behind publication because harmonised international collections close their cycle slowly.

Granularity is deliberately national. Every row is a country sliced by ISCED 2011 level, sex, age band and unit of measure - no sub-national breakdowns in most tables. That constraint is also the guarantee: the same grain everywhere is what makes Germany's 91.7 percent of three-year-olds comparable with whoever's number you put next to it.

Against the OECD's education flows, this is the tighter but deeper-on-Europe view: the OECD adds partner economies and institutional mechanics such as class size and instruction time, while Eurostat holds the candidate-country bridge and the expenditure-to-GDP ratios the Commission actually governs by. Against UNESCO's UIS browser, trade geographic breadth for definitional strictness - UIS reaches nearly every country on earth at coarser harmony, this stops at Europe's edge but enforces one standard harder.

Practical pairing: use these tables as the harmonised backbone for anything EU-facing, then reach sideways when a question leaves the frame. Enrolment trends and early-leaver rates stay here; sub-national school detail moves to a national source; global benchmarking borrows UIS or EdStats.