data.gouv.fr — French national education & schooling datasets

It is the education slice of data.gouv.fr, the French state's official data catalogue run by Etalab: roughly 354 datasets contributed by ministries, regions, departments and municipalities, all reachable under a single tag.

The anchor is Annuaire de l'éducation, compiled by the Ministères de l'Éducation nationale from ONISEP and RAMSESE records. This is the register of every public and private school and institution operating in France - first- and second-degree schooling, medico-social establishments, CIO information centres and administrative bodies - currently 68,940 rows described across 71 columns. Because it is a register rather than a survey, it answers presence and provision questions (does this commune have a collège with a canteen?) that no statistical table can.

Around it sit the analytical layers: Ministry-published Parcoursup admission files cut by speciality, INSEE's detailed FOR1 schooling statistics, and hundreds of school-location datasets published by sub-national governments. Four layers, one catalogue - national register, admissions, statistics, local geography.

One row per establishment, wide format. Row 0301025B is Ecole élémentaire Georges Lapierre in Laudun-l'Ardoise (30290): an Ecole, publicly run, department 030, académie 11, région 76. Its remaining ~60 columns are mostly 0/1 switches recording what the school offers - elementary level, canteen, specialty sections - plus the geocode triple and lifecycle stamps.

That flatness is the point. Nothing requires interpretation: the public/private switch is a column, the REP flag is a column, boarding is a column. Analysts coming from survey data usually notice two things - first, that the UAI code makes every row permanently joinable against other ministry publications; second, that precision_localisation grades the geocode rather than pretending all coordinates are equal.

The other layers break the pattern deliberately. Parcoursup rows report programmes and admissions rather than buildings; INSEE FOR1 reports pupil counts by commune and level. Sample rows above show attributes observed in the catalogue this session; full row-level extracts ship with your sample.

The dictionary covers the directory spine: Identifiant_de_l_etablissement (the UAI code), identity and status columns, the address and administrative-geography stacks, the provision flags, Appartenance_Education_Prioritaire, SIREN_SIRET, the geocode triple, and the lifecycle fields date_ouverture, date_maj_ligne and etat.

These definitions are verified against live directory records rather than inferred from a layout document, which matters because the wider catalogue has no unified schema - each contributor documents itself, and local school-location datasets in particular vary in column naming. We pin per-file schemas when cutting your sample so nothing surprises you at mapping time.

Fields outside the directory - Parcoursup speciality variables, FOR1 measures - are documented separately on request with your sample.

Geographically: metropolitan France plus overseas territories, down to the individual establishment - the finest grain in the catalogue. Aggregated products climb from commune through department and académie to national. If your frame is cross-country comparison, this catalogue gives you the French side in unusual depth and stops at the border; harmonised European tables slot in above it.

Temporally, the layers work differently by design. The directory behaves as a register: rows carry opening dates, revision stamps and an operational status, so closures, mergers and reopenings are events in the data rather than corrections to it. INSEE statistical series are annual. Parcoursup files accumulate per campaign year, which suits cohort analysis better than a rolling snapshot would.

Grain follows purpose. Establishment rows answer where-provision questions; commune-to-national aggregates answer system-state questions. Teams rarely need just one, which is why the whole tag ships as a single deliverable.