World Bank Education Statistics (EdStats/WDI) — Harmonized Education Indicators

Two complementary collections under one roof. The larger is the EdStats compilation - 8,450 series assembled from the UNESCO Institute for Statistics, the OECD and the World Bank's own collections, reaching across school access and completion, learning outcomes (including harmonized adjusted test scores), teacher stocks, education expenditure and attainment. The smaller is the education topic inside the World Development Indicators: 156 indicators carrying the SE.* prefix, covering enrolment, literacy, progression, pupil-teacher ratios and education spending as a share of GDP.

The division of labour matters less than the shared grammar. Both collections key observations on the same country codes and the same indicator-code conventions, so a series pulled from either shelf reconciles against the other without translation work - a rarer property than it sounds in cross-country education data, where competing compilations routinely disagree about whether a figure covers a calendar or a school year.

One honest caveat belongs up front: the two shelves age differently. Some UIS-sourced series in the EdStats compilation carry an earlier end-year than their WDI counterparts, so a join across shelves deserves an end-year check rather than an assumption.

Tidy and long-format: one observation per indicator, country and year, each row self-describing through the attribute set in the dictionary below. Two observed rows make the texture concrete. India's primary enrolment reads 138,368,073 pupils for 2024 and 126,117,112 for 2025 - a drop worth noticing rather than smoothing over, because the newest year of a reporting series can arrive partially complete while returns are still collected. Treat trailing-edge years as provisional until you have checked them.

Nulls are statements too. The EdStats adjusted primary math score for Brazil at reference year 2095 ships as a genuine null: the projection framework reserves the horizon even where no value has been published. Filtering those rows out of historical analysis takes one predicate, provided you know they exist - which is exactly the kind of thing a sample catches before a pipeline does.

Every observation also carries obs_status, the flag marking estimates and missing-data conditions explicitly. Between labelled nulls and flagged estimates, the data distinguishes between 'zero', 'unknown' and 'not yet published' - a distinction most spreadsheets flatten at your cost.

Nine attributes, four blocks. Identification: indicator.id and indicator.value. Geography: country.id/country.value and countryiso3code. Period: date. Measure: value, unit, decimal and obs_status. The full definitions, with examples drawn from live observations, sit in the table above.

The design choice worth appreciating is redundancy in the geography block. Carrying both a two-letter and a three-letter country code per row looks wasteful until the day your CRM keys on one and your analytics warehouse on the other - then it is the difference between a join and a mapping project.

Because this attribute grammar is identical on both shelves, the dictionary above covers the entire surface: the 156 SE.* WDI indicators and the 8,450 EdStats series speak the same row shape. New indicators extend the catalogue without changing the container.

Geographically: 217 economies, plus the World Bank's regional and income aggregates - 272 selectable entities once those are counted. The aggregates are more useful than decorative; comparing East Asia & Pacific against Europe & Central Asia on spending shares is one filter rather than a hand-built country list.

Temporally the shelf runs deep. WDI education series reach back to 1970, giving half a century of annual history on enrolment, progression and spending - long enough to watch demographic waves move through school systems. The EdStats side adds forward projections with dated values out to 2100, so history and forecast share one column and separate cleanly on the date alone.

Grain is the hard boundary: annual, national, full stop. There are no subnational cuts here. If the question needs districts, schools or classrooms, this shelf answers the country-frame question and a national source answers the local one - teams that expect otherwise are the reason we pin scope before cutting samples.

Think in rings. This is the widest ring: 217 economies, annual, indicator-level. The OECD Education at a Glance collection narrows the ring to OECD members and partners but compensates with deeper metadata slices - ISCED levels, public/private splits, earnings premia. **Eurostat's educ_* tables** go narrower still, trading breadth for EU harmonization depth.

Against national sources the relationship flips: the US NCES compilations and England's EES platform trade coverage for grain, running to state, district or school level where this shelf stops at the border. The natural architecture uses this collection as the frame and national sources as insets.

If the live question is institutional - how a US-centric postsecondary series compares against this global frame - we have put the two head-to-head on the comparison page.