Brazil’s regional gap is not the poverty gap in disguise, and it is not the rural gap either

The North and Northeast sit far below the Southeast. Holding socio-economic status constant removes about a third of the distance; holding urban and rural composition constant as well removes almost none of the rest.

Dr Milos Kankaras | ORCID 0000-0002-3190-7751 | Center for Psychology, Podgorica | PISA 2022 country notes, Latin America | August 2026

Download the full note (PDF)

Summary

Brazil’s regional inequality is the oldest fact in Brazilian education policy, and the most common way of dismissing it is to say that it is really income inequality with a map drawn on it. That is a testable claim and this note tests it.

Brazil is also one of the few PISA participants with adjudicated regional estimates, so the five macro-regions are supported by the sampling design rather than being a slice of a national sample that happens to have enough cases.

What the data shows.

• The North scored 31.2 points below the Southeast (5.4) and the Northeast 25.8 points below (4.4). Between the Northeast and the South, the two extremes of the distribution, the distance is roughly 31 points.
• Socio-economic status accounts for about a third. Net of the PISA index the North sits 22.6 points below the Southeast and the Northeast 12.9 points below.
• Urban and rural composition accounts for almost nothing further. Holding school location constant as well, the North is 23.2 points below the Southeast (4.4) and the Northeast 11.2 points below.
• The rural gap is separately real and separately large: village and rural schools sit 39.6 points below city schools, 23.0 net of socio-economic status. It simply is not what makes the North and Northeast different.

What survives both adjustments is a difference between comparable students, in comparable settlements, at comparable socio-economic levels, in different parts of the country. That is the part a federal equalisation policy could in principle move, and it is roughly twenty points.

Why this is the question for Brazil

Brazil runs one of the world’s largest federal education equalisation systems, and the case for it rests on the claim that regional differences are not merely a reflection of where poor people live. If the regional gap disappeared under socio-economic adjustment, the honest conclusion would be that the instrument should be targeted at households rather than at states. It does not disappear.

The second explanation worth eliminating is settlement pattern. The North and Northeast are more rural, rural schools perform worse everywhere, and a regional gap that was really a rural gap would call for a different instrument again. That explanation is also tested here, and it also fails.

Two eliminations do not identify a cause. They do narrow the field considerably, and they are the kind of result that a national study centre can act on.

The data, and how it was handled

10,798 students in 598 schools, representing a weighted population of 2,262,972 15-year-olds, from the OECD’s PISA 2022 student and school Public Use Files. National mean in mathematics: 378.7 points (SE 1.6).

Achievement is imputed rather than measured, so estimates are pooled by Rubin’s rules over all ten plausible values. The sample is clustered in schools, so standard errors come from the 80 Fay-adjusted replicate weights the OECD ships for the purpose. Omitting either understates the uncertainty, usually by a factor of two or more.

Source: OECD, PISA 2022 Results (Volume I and II) Country Note: Brazil, published 5 December 2023, https://www.oecd.org/en/publications/pisa-2022-results-volume-i-and-ii-country-notes_ed6fbcc5-en/brazil_61690648-en.html (read 6 August 2026)

Five regions, two adjustments

The five macro-regions with their socio-economic profiles alongside performance. The Southeast is the reference throughout because it is the largest, at 40.5% of the cohort.

The socio-economic ordering and the performance ordering are similar but not identical, which is the first hint that one does not reduce to the other. The Middle-West and the South sit at almost the same socio-economic level as the Southeast and perform differently from each other.

Adding school location to the adjustment is where the rural explanation is tested, and it barely moves the coefficients. Between two students of the same socio-economic background in the same type of settlement, one in the North and one in the Southeast, roughly 23.2 points remain.

The South is the mirror case and is worth a sentence. It performs above the Southeast despite a slightly lower socio-economic profile, and its advantage grows to 8.9 points under adjustment. Any explanation of Brazil’s regional pattern has to account for the South as well as for the Northeast, and most do not try.

Where students are in the system

Grade position and programme type separate Brazilian 15-year-olds more sharply than region does, and the vocational figure is the one that surprises people.

Students in upper secondary vocational programmes outperform those in general upper secondary, which is the reverse of the pattern in most European systems and reflects who gains access to vocational places in Brazil rather than what those places do. The 18.0% still below upper secondary at 15 are the group carrying the largest disadvantage, at 57.9 points net of socio-economic status.

Can this comparison be trusted? A measurement audit

Test scores are placed on a common scale by design. Questionnaire indices are not. Comparing an index across groups assumes the items mean the same thing in every group being compared, and that assumption is testable. It is usually not tested.

Teacher support in mathematics is the questionnaire construct most often invoked in Brazilian regional debate, usually to argue that the North and Northeast suffer from weaker teaching. Comparing that index across regions presumes the items mean the same thing in Belem and in Sao Paulo. Before making the comparison, the presumption is tested.

Measurement invariance of the teacher support in mathematics block across region. N = 8,321. Estimator MLR. Decision rule after Chen (2007).

The teacher-support block behaves unusually well: configural fit is excellent, with CFI near 0.999 and RMSEA near 0.035, and the cascade reaches full invariance under the continuous estimator and scalar invariance under the categorical one. Of the eight constructs tested across this series, this is the cleanest.

So the regional comparison of reported teacher support is licensed, and it can be used in the Brazilian regional debate without the caveat that attaches to most questionnaire indices. That is worth knowing precisely because it is not the usual result.

What follows

Three statements are supported and narrow enough to defend.

• Brazil’s regional gap is not reducible to socio-economic composition. About a third of the raw North and Southeast difference is socio-economic; roughly 23.2 points survive both socio-economic and urban-rural adjustment.
• It is not the rural gap either. The rural gap is separately real at 23.0 points net of socio-economic status, and controlling for it leaves the regional coefficients essentially unchanged.
• The South, not only the Northeast, is the anomaly requiring explanation: it outperforms the Southeast from a slightly lower socio-economic base.

The natural next analysis is a school-level decomposition within region, separating between-school from within-school variance. Brazil’s between-school variance is among the highest in PISA, and a regional gap that lives between schools implies a different instrument than one that lives inside them. That requires a multilevel model, which is a deep-dive rather than a note.

What this note does not claim

• No trend statement. Comparing 2022 with an earlier cycle requires the published link error for that cycle pair to be carried in the variance. No such comparison is made here.
• No causal claim. Every difference is an association measured at one point in time. “Net of ESCS” means one measured index is held constant, not that other things are equal.
• The invariance cascade is fitted without the replicate-weight design, which is standard practice and is stated rather than left implicit.
• Regional estimates are adjudicated for Brazil, so they are properly supported, but the macro-regions are internally heterogeneous to a degree that a five-way split cannot show. State-level analysis is not supported by the public sample.
• The vocational finding reflects selection into vocational places and should not be read as evidence about the effect of vocational programmes.

Method and reproducibility

Data: OECD PISA 2022 Public Use Files, downloaded from webfs.oecd.org on 6 August 2026. Point estimates pooled over ten plausible values by Rubin’s rules; sampling variance from 80 Fay-adjusted replicate weights with a Fay factor of 0.5, computed per plausible value and averaged; imputation variance inflated by (1 + 1/M). Invariance tested as a staged configural, metric, scalar and strict cascade in lavaan with full-information estimation for the rotated questionnaire design, plus a categorical sensitivity re-run on pairwise-present data.

Subgroups are derived from variables the OECD codes identically in every participating system, so this note was produced without country-specific recoding. Where a country’s own sampling strata carry a more policy-relevant structure, that structure is used instead and the note says so.

Enquiries about reproducing the analysis are welcome at milos@centerforpsychology.me. Commissioning the equivalent for another country is consulting work, handled by AdriaMont Consulting DOO at milos@adriamont.me.

All country notes in this series