OECD Education at a Glance & OECD Data Explorer — Education topic

One catalogue, four shelves. The institutional organisation shelf measures how compulsory schooling actually runs: intended instruction time by age, class size, school type - the mechanics most international comparisons skip. The finance shelf carries the UOE expenditure series, splitting public and private sources and current versus capital spending per student. The people shelf counts them: enrolment and participation by ISCED level, vocational programme distribution, graduation rates, attainment, teacher salaries benchmarked against comparable workers.

The fourth shelf is the one competitors forget: outcomes. The education-and-earnings flow tracks relative earnings by attainment level, education field, sex, age band and labour-force status, reaching roughly a million rows back to 2013 - enough history to watch the graduate premium move within cohorts rather than assert it from one year's snapshot. Alongside sit TALIS survey outputs on teachers and principals, the profession's only genuinely comparative self-portrait.

Scale, concretely: 1,546 dataflows live in the OECD's statistical warehouse, 141 of them belonging to the education agencies, and a keyword sweep surfaces 455 education-related series once thematic collections are counted. That is not a spreadsheet with delusions - it is a system of records with versioned definitions.

A row is an observation plus its complete provenance. One from the institutional flow reads: Australia, public institutions, intended instruction time measured as a percentage of total compulsory instruction, all ISCED levels combined (_Z), eighteen-year-olds, language subjects, reference year 2025 - and an empty value with an M status flag, because that cell was never reported and the data says so instead of hiding it. Another, from the earnings flow: Slovenia, women aged 45 to 54, ISCED 5 attainment, employed, relative earnings, 2013.

That structure is the whole point. Because dimensions are columns rather than file conventions, the same query pattern pulls instruction time in Japan and class size in Chile without anyone re-parsing anything. Country codes stay stable across all 141 flows, which turns 'compare 38 education systems' into a group-by rather than a translation project. And because every value drags its status flag along, gaps are auditable - you can count how much of a table is real before building an argument on it.

The sample block above shows dimension combinations exactly as indexed this session; full observations with values arrive with your sample, pinned to the specific flows you name.

The spine above is verified against live records: dataflow pins the row to its indicator set and version; ref_area locates it geographically; measure and unit_measure state what was counted and in what units; inst_type_edu, education_lev, age and subj_type slice it; obs_value and obs_status deliver the number and its caveat; ref_period stamps the year. The earnings flow swaps in time_period and adds sex, attainment_lev and labour_force_status. unit_mult handles scale factors so a value expressed in thousands reads correctly next to one expressed in units.

Two fields earn their keep more than any others. unit_measure is where careless cross-country comparisons die - a percentage-of-total-instruction figure and an hours-per-year figure look identical until you read the unit. And obs_status converts silence into information: M means missing, not zero, and treating it otherwise is how bad dashboards get built.

Flows beyond the ones sampled here carry their own dimension lists - VET distributions add programme orientations, TALIS surveys use instrument-specific variables. Those schemas ship documented with your sample rather than guessed at here.

Geographically this is the widest lens in the industry's education shelf: all 38 members, from Australia to the United States, plus partner economies that let you benchmark emerging systems against established ones. Selected flows push below the national border - one institutional flow alone resolves 137 distinct reference areas, including Canadian provinces and other TL2 territories, so a provincial ministry can be compared directly against national averages elsewhere.

Temporally the flows refuse to synchronise, and that asymmetry is informative. Institutional indicators expose recent reference years - 2024 and 2025 in the current edition - because they describe the running system. The earnings flow stretches back to 2013 with about a million rows, because cohort outcomes need runway before they mean anything. Education at a Glance editions date to 1992, giving three decades of context for any trend line you draw. Check per-flow spans before committing to a series design; assuming uniform depth across 141 flows is the fastest way to build a chart with holes.

Grain follows purpose. System-state questions resolve at country-by-year, sliced by ISCED level and institution type. Individual-outcome questions resolve through the demographic cuts - sex, age band, attainment, field, labour-force status - without ever exposing person-level records. Both ends of that spectrum live under one keying scheme, which is why analysts who start with one flow rarely stop there.

Three neighbours matter. **Eurostat's educ_* tables cover EU members with tighter harmonisation but a shorter geographic reach - no Japan, Korea or partner economies. UNESCO's UIS browser goes wider than anyone globally but trades depth for breadth; fine-grained finance splits thin out fast outside the OECD. World Bank EdStats** harmonises across income levels, useful for development framing, at the coarsest grain of the three.

This dataset's edge is the pairing of width with depth: partner economies plus partner-level detail, thirty-plus years of editions, and outcome measures - especially the earnings-by-field-and-gender cuts - that none of the three replicate at comparable resolution. Its cost is the SDMX convention itself: dimensions-as-columns rewards teams comfortable with codelists and punishes anyone expecting tidy wide tables pre-wrapped. If your analysis is EU-only, Eurostat will feel lighter. If it touches any OECD economy outside Europe, or any earnings question, this is the deeper well.