Reading gaps become misleading when they are assigned to language alone. The strongest international evidence supports a benefit from linguistic familiarity while also showing that social background and school capacity can radically change the size of that benefit. [OECD 2023 · 2375] [OECD 2023 · 2377]
Key findings
The most consistent signal across the evidence is that comprehension of the language used for schooling removes a genuine obstacle to learning to read. Yet the size of the observed advantage is inseparable from family resources, migration history and the quality of schools. PISA 2022 is especially useful because it shows how a large immigrant reading gap changes when socioeconomic background and home language are considered rather than treated as invisible differences. [OECD 2023 · 2375] [UNESCO 2025 · 2359]
The practical answer is conditional. Familiar language generally supports reading, but a high alignment rate does not automatically produce high achievement or low learning poverty. Strong school systems can also compensate for widespread mismatch when they provide structured second-language instruction. The decisive question is not only which language is named in policy, but how long it is sustained, how the additional language is introduced, and whether teachers and materials make the transition workable. [World Bank 2021 · 2374]
Four indicators answer four different questions
UIS indicator 4.5.2 measures the share of enrolled students whose first or home language is the language of instruction. Indicator 4.1.1 concerns minimum proficiency in reading at defined education stages, while 4.1.2 is a completion-rate measure for a cohort several years above the intended age of the final grade. These indicators can be connected analytically, but they are not interchangeable measures of the same outcome. [UNESCO Institute for Statistics 2025 · 2385] [UNESCO Institute for Statistics 2026 · 2387] [UNESCO Institute for Statistics 2026 · 2354]
Learning poverty adds another layer because it combines learning deprivation with schooling deprivation around end-primary age. A language policy could improve the reading of enrolled children while the learning-poverty rate remains high if many children are still out of school. Secondary completion reflects an even longer chain of decisions and constraints. Keeping the outcomes separate is essential if the report is to distinguish a classroom mechanism from a broader education-system result. [UNESCO Institute for Statistics Learning Data Resource Center 2026 · 2386] [World Bank / UNESCO Institute for Statistics 2026 · 2388]
Interpretation also requires separating the direction of an association from its magnitude. The idea that a comprehensible language eases learning is repeated across datasets and guidance, but a point difference observed in one system cannot be transferred mechanically to another. Home language, assessment language, official language of instruction and the language a teacher actually uses can all be different variables. [UNESCO Institute for Statistics 2025 · 2385] [OECD 2023 · 2389]
The OECD-average reading gap falls from 39 points to 9
In PISA 2022, the raw OECD-average reading difference between immigrant and non-immigrant students was 39 points. After economic, social and cultural status was taken into account, the gap was 25 points. When home language relative to the assessment language was considered as well, the difference fell again to 9 points. Few available statistics show the overlap among migration, social background and language this clearly. [OECD 2023 · 2375]
Those steps should not be read as a mechanical decomposition in which one variable “causes” a fixed number of points. The controls are correlated and the sequence of adjustment is not a causal experiment. What can be said with confidence is that the raw immigrant gap exaggerates what can reasonably be attributed to language, while the further reduction after adding home language also shows that language exposure cannot simply be discarded. [OECD 2023 · 2389]
Italy and Germany show why one adjusted average is not enough
Italy offers a case where the observed gap becomes much weaker once social background is taken seriously. Among immigrant students, 59% reported a home language different from the assessment language and the raw reading gap was 31 points. After adjustment for students’ socioeconomic profile, the difference was no longer statistically significant. [OECD 2023 · 2376]
Germany points in another direction. Sixty-three per cent of immigrant students reported a different home language, the raw reading gap was 67 points, and 40 points remained after socioeconomic adjustment. The residual may reflect language proficiency, age at migration, tracking, peer composition or differences in school support. Together, the two countries rule out the simple claim that socioeconomic adjustment always makes language-related gaps disappear. [OECD 2023 · 2377]
| Comparison | Raw reading gap | After SES | After SES + language | Interpretation |
|---|---|---|---|---|
| OECD average | 39 | 25 | 9 | Both SES and home language matter |
| Italy | 31 | Not significant | — | Gap disappears after SES adjustment |
| Germany | 67 | 40 | — | Large gap remains after adjustment |
Figure 1 is a picture of heterogeneity, not a country ranking
The first figure puts the OECD adjustment sequence beside the Italian and German country results. Its purpose is to show how much the interpretation changes when background variables are added. The OECD average moves into single digits when home language is included, while Germany still has a large gap after socioeconomic adjustment. [OECD 2023 · 2375] [OECD 2023 · 2377]
PISA’s home-language variable is also a reminder about measurement. It records whether the language mainly spoken at home matches the assessment language, not the full language ecology of classroom instruction. The chart therefore cannot be used as a direct causal estimate of language-of-instruction policy. It is best read as evidence about how language, migration and socioeconomic conditions overlap within student populations. [OECD 2023 · 2376] [OECD 2023 · 2389]
Timing matters as much as definition. Early-grade decoding, end-primary minimum proficiency, PISA reading at age 15 and secondary completion are outcomes at different stages. Demonstrating that an early language intervention changes later completion would require longitudinal cohort evidence. International datasets are much stronger for cross-sections than for that long causal chain. [IEA / Boston College 2023 · 2390] [UNESCO Institute for Statistics 2026 · 2353]
Singapore shows that mismatch can coexist with very high achievement
Singapore is a strong counterexample to the idea that a home/assessment-language mismatch necessarily predicts weak reading. In PISA 2022, 64% of immigrant students and 28% of non-immigrant students reported speaking a different language at home. Yet immigrant students held a raw 12-point reading advantage, and the adjusted difference was not statistically significant. [OECD 2023 · 2381]
The institutional context helps explain why mismatch is not destiny. English is the main medium of schooling, while students also study a designated Mother Tongue Language such as Chinese, Malay or Tamil. That arrangement does not eliminate the effort of learning through a language that may differ from home use, but it demonstrates that a well-supported bilingual system can manage the transition without producing a simple mismatch penalty. [Singapore Ministry of Education 2023 · 2407]
South Africa shows that alignment alone cannot prevent a reading crisis
South Africa provides the opposite warning. In PIRLS 2021, 66% of Grade 4 learners reported that they always spoke the test language at home. Nevertheless, 81% were below the international low benchmark of 400, and the national Grade 4 reading score fell from 320 in 2016 to 288 in 2021. [South Africa Department of Basic Education 2023 · 2382] [South Africa Department of Basic Education 2025 · 2368]
This does not mean language alignment is irrelevant. It means alignment is not sufficient when pandemic disruption, poverty, scarce reading materials, weak instructional capacity and uneven school functioning are large. South Africa’s 2025 rollout of mother-tongue-based bilingual education beyond Grade 3 can be read as an attempt to sustain language support longer, but its success will still depend on the broader conditions that shape teaching quality. [South Africa Department of Basic Education 2023 · 2410]
The long-term shift is from short mother-tongue phases to planned multilingual pathways
UNESCO’s education-language work dates back to the 1950s, and by 1999 multilingual education was framed around the use of a mother tongue together with regional or national and international languages. The 2003 position paper paired the educational case for mother-tongue learning with the need to give learners access to languages of wider communication. [UNESCO 2025 · 2370] [UNESCO and Education 2030 partners 2015 · 2364] [Global Education Monitoring Report Team, UNESCO 2025 · 2360]
After 2015, the debate became more tightly linked to SDG4 measurement and learning poverty. The World Bank estimated in 2021 that about 370 million students in low- and middle-income countries—roughly 37%—were learning in a language other than the one they understood best. UNESCO’s recent guidance goes beyond simply choosing the home language: it emphasises a planned multilingual route, potentially maintaining it for 6-8 years while another language is added gradually with trained teachers and suitable materials. [UNESCO 2003 · 2371] [World Bank 2021 · 2374]
Learning poverty and minimum reading proficiency should not be collapsed into one outcome
Learning poverty combines two problems: not reaching the minimum reading threshold and being out of school. This means progress in classroom language support may improve the first component without changing the second. It also means a country can change its learning-poverty rate through access and retention even when instructional language has changed little. [World Bank 2024 · 2355] [UNESCO Institute for Statistics 2026 · 2353]
The World Bank learning-poverty database covers more than 115 countries, making it attractive for cross-country analysis. But indicator 4.5.2 has a narrower set of observed country-years. Requiring the same countries and years for language alignment and learning poverty can reduce the sample sharply. A precise-looking global coefficient would therefore risk being driven by a small, selected set of countries rather than by a truly global panel. [World Bank / UNESCO Institute for Statistics 2026 · 2388]
Learning-poverty analysis requires an additional distinction between pathways. Better classroom comprehension can raise minimum reading proficiency, while expanded access can reduce the out-of-school component of learning poverty. If both improve at once, attributing the combined change to language policy requires data that follow access and achievement separately rather than treating the composite indicator as one mechanism. [UNESCO Institute for Statistics 2025 · 2385] [OECD 2023 · 2389]
Secondary continuation is plausible as a pathway, but the direct quantitative link is weaker
The mechanism connecting language comprehension to persistence is credible: students who cannot follow instruction can fall behind, repeat grades and disengage. UNESCO’s GEM work has long connected language exclusion among minority and Indigenous learners with both learning and continuation. Recent multilingual-education guidance also links a sustained familiar-language pathway with lower repetition and dropout risk. [UNESCO Global Education Monitoring Report 2016 · 2372] [UNESCO Institute for Statistics 2026 · 2387]
Completion, however, is a cumulative outcome shaped by costs, distance, adolescent work, gender norms, migration and conflict as well as school learning. UIS provides a comparable 4.1.2 completion measure, but there are not enough matched observations with language-of-instruction alignment to identify a stable independent cross-country effect. The evidence is therefore stronger on the mechanism and on selected programmes than on a single average completion coefficient. [Global Education Monitoring Report Team, UNESCO 2025 · 2360] [UNESCO Institute for Statistics 2026 · 2363]
What remains after wealth, location and sex are considered
International data are increasingly capable of moving beyond national averages. The World Bank’s harmonised learning-outcome dataset provides breakdowns by sex, urban-rural location and wealth quintile. UNICEF’s MICS Foundational Learning Skills module directly assesses children aged 7-14 and can include out-of-school children; in MICS6 it covered 46 countries or territories, around 230,000 children and more than ten assessment languages. [World Bank 2024 · 2356] [UNICEF DATA 2026 · 2378] [IEA / Boston College 2023 · 2390]
These data can test whether lower achievement among language-mismatched learners is mostly a reflection of poverty or rural disadvantage. But levels of analysis must remain separate. UIS language-alignment indicators are national aggregates, while PISA and PIRLS support student-level models. Pooling them mechanically in one regression can create an ecological fallacy. A safer design estimates relationships at each level and then checks whether conclusions converge. [UNICEF DATA 2026 · 2362] [OECD 2023 · 2357]
| Dataset | Main unit | Language variable | Socioeconomic breakdown | Best use |
|---|---|---|---|---|
| UIS SDG4 | Country-year | Actual LoI alignment (4.5.2) | Sex/location/wealth where available | Link alignment to proficiency/completion |
| PISA 2022 | 15-year-old student | Home vs assessment language | ESCS/sex/immigration | Student-level adjustment |
| PIRLS 2021 | Grade 4 student | Home/test-language context | Home and school resources | Primary reading comparisons |
| MICS FLS | Child/household age 7-14 | Survey-language context | Urban-rural/wealth/sex | Foundational learning incl. out-of-school |
| Learning Poverty | Country/assessment year | No direct language variable | Country-level | Combine reading deprivation and schooling |
Counterevidence and alternative explanations
The first alternative explanation is selection. Learners who use a different home language may also be disproportionately recent migrants, low-income students or members of disadvantaged communities. The OECD average and the Italian result show how much of a raw language-related gap can shrink once those characteristics are considered. [OECD 2023 · 2375] [OECD 2023 · 2381] [UNESCO Institute for Statistics 2025 · 2385]
A second explanation is compensating system capacity. Singapore illustrates how high mismatch can coexist with excellent results when language instruction and schooling are strong. A third is measurement. Speaking the test language at home is not the same as understanding complex classroom discourse, and official language-of-instruction rules may differ from teachers’ real code-switching practices. South Africa warns against using an alignment percentage as a proxy for school quality. [OECD 2023 · 2376] [South Africa Department of Basic Education 2023 · 2382]
Scenarios and practical implications: when should alignment have the largest effect?
The most favourable scenario combines strong language support with strong system support. Learners build literacy in a language they understand, an additional language is introduced gradually, teachers are capable of teaching across the transition, and materials are available in the subjects students actually study. This is why recent guidance treats duration, teacher preparation and resources as part of language policy rather than as optional add-ons. [Global Education Monitoring Report Team, UNESCO 2025 · 2360] [OECD 2023 · 2381]
Where system support is strong, mismatch can sometimes be compensated for, as Singapore suggests. Where system support is weak, alignment by itself may not be enough, as the South African evidence shows. Monitoring should therefore ask whether minimum reading proficiency, inequality, attendance or repetition, completion, teacher readiness and material supply improve together—not only whether an alignment rate moved upward. [UNESCO 2025 · 2359] [South Africa Department of Basic Education 2023 · 2382]
The conditions matrix also needs a time dimension when it is used for policy evaluation. New language programmes can initially face shortages of trained teachers and materials, while benefits from stronger foundational literacy and gradual second-language acquisition may emerge only after several cohorts. One-year achievement changes are therefore a weak basis for declaring a transition successful or unsuccessful. [IEA / Boston College 2023 · 2390] [UNESCO Institute for Statistics 2026 · 2353]
A derived matrix for interpreting language-of-instruction evidence
Putting the evidence together produces four useful conditions. High language support combined with high system support is the most consistently favourable. Low language support can be partly compensated for by a strong school system. High language support with weak system capacity may still leave large learning deficits. When both forms of support are weak, comprehension demands and socioeconomic disadvantage are more likely to accumulate. [OECD 2023 · 2381] [UNESCO 2025 · 2359]
This is a qualitative interpretation matrix, not a ranking of countries. It is derived from the contrast between cases such as Singapore and South Africa and from guidance that repeatedly joins language choice to teacher capacity and materials. A country should be placed in a cell only after actual classroom language, resource conditions and learner outcomes have been verified. [South Africa Department of Basic Education 2023 · 2382] [Global Education Monitoring Report Team, UNESCO 2025 · 2360]
Limitations of the available evidence
The largest limitation is mismatch in years and units of observation. UIS indicator 4.5.2 is closest to actual language-of-instruction alignment but has restricted coverage. PISA and PIRLS are much stronger for student-level adjustment, yet their home- and assessment-language variables do not fully observe the language used by a teacher across subjects and grades. Learning poverty incorporates children outside school but does not measure language exposure directly. [UNESCO Institute for Statistics 2025 · 2385] [IEA / Boston College 2023 · 2390] [Zambia Ministry of Education 2024 · 2367]
Implementation also differs from policy text. India’s NEP recommends the home, mother or local language as the medium through at least Grade 5 and preferably Grade 8 where possible, but practice varies across states and schools. Zambia has moved in the other direction, shifting from regional-language instruction toward English across levels while encouraging code-switching in foundational learning. A policy date is therefore a poor substitute for measured exposure. [OECD 2023 · 2389] [Government of India, Ministry of Education 2020 · 2365]
Conclusion: align language support with the rest of the school system
The international evidence from 2015-2025 supports a more specific conclusion than “mother tongue is better.” Comprehensible language reduces a real learning barrier, and sizable gaps can remain even after socioeconomic adjustment in some systems. But other systems see gaps disappear after adjustment, high mismatch can coexist with high achievement, and substantial alignment can coexist with a reading crisis. [OECD 2023 · 2375] [OECD 2023 · 2381] [Global Education Monitoring Report Team, UNESCO 2025 · 2360]
Reducing inequality therefore requires more than raising an alignment percentage. Duration, second-language transition, teacher capability, materials and family background have to be treated as one design problem. A stronger estimate of the independent effect on learning poverty and secondary completion will require matched UIS, World Bank and student-level data. For now, the most defensible answer is that familiar language helps most when the school system is capable of turning comprehension into sustained instruction. [OECD 2023 · 2377] [South Africa Department of Basic Education 2023 · 2382]
Sources and method
The principal comparisons draw on PISA 2022 results on immigrant background, home language and socioeconomic status, with PIRLS 2021 and South African official material supplying primary-grade context. Definitions come from UIS indicators 4.5.2, 4.1.1 and 4.1.2 and the World Bank learning-poverty metadata. Wealth, location and out-of-school dimensions are informed by the harmonised country learning-outcome dataset and UNICEF MICS. [OECD 2023 · 2375] [South Africa Department of Basic Education 2023 · 2410] [World Bank / UNESCO Institute for Statistics 2026 · 2388] [UNICEF DATA 2026 · 2362] [South Africa Department of Basic Education 2025 · 2368] [Brazil Ministry of Education 2025 · 2369]
Numbers were not combined when their units were different. Home-versus-assessment-language mismatch was not treated as identical to classroom language-of-instruction mismatch, and national aggregates were not pooled directly with student microdata. Policy evidence from India, South Africa, Zambia and Brazil was used to interpret implementation conditions rather than to manufacture a cross-country causal coefficient. [International Association for the Evaluation of Educational Achievement (IEA) 2023 · 2358] [UNESCO Institute for Statistics 2025 · 2385] [World Bank 2024 · 2356] [Government of India, Ministry of Education 2020 · 2365] [Zambia Ministry of Education 2024 · 2367]
Selected public sources
- OECD — Immigrant background and student performance: PISA 2022 Results (Volume I)
- OECD — PISA 2022 country note — Italy
- OECD — PISA 2022 country note — Germany
- OECD — PISA 2022 country note — Singapore
- South Africa Department of Basic Education — PIRLS 2021 South African Main Report
- South Africa Department of Basic Education — DBE briefing on PIRLS 2021 reading literacy
- World Bank — Loud and Clear — effective language of instruction policy approach
- UNESCO Institute for Statistics — UIS Data Browser — SDG 4 monitoring data
- UNESCO Institute for Statistics — SDG 4 Indicators — Target 4.5 and Target 4.1
- World Bank — Learning Poverty Global Database: Historical data and sub-components
- World Bank — Global Dataset of Country Learning Outcome
- OECD — PISA 2022 Database
- International Association for the Evaluation of Educational Achievement (IEA) — PIRLS 2021 International Database
- UNESCO — Languages matter: Global guidance on multilingual education
- Global Education Monitoring Report Team, UNESCO — The right multilingual education policies can unlock learning and inclusion
- UNICEF DATA — The Foundational Learning Skills Module: Leveraging child-centric surveys to measure learning outcomes
- UNESCO Institute for Statistics — SDG 4 scorecard progress report on national benchmarks: focus on completion rates, 2026
- UNESCO Institute for Statistics — Indicator 4.5.2 — Percentage of students whose first/home language is the language of instruction
- UNESCO Institute for Statistics Learning Data Resource Center — Minimum proficiency levels for SDG 4.1.1
- UNESCO Institute for Statistics — Completion rate — UIS glossary
- World Bank / UNESCO Institute for Statistics — Learning Poverty metadata — SE.LPV.PRIM
- OECD — PISA 2022 — construction of reporting scales and indices
- IEA / Boston College — PIRLS 2021 International Database and User Guide
- Government of India, Ministry of Education — National Education Policy 2020
- Zambia Ministry of Education — 2023 Zambia Education Curriculum phased implementation announcement
- South Africa Department of Basic Education — Mother Tongue-based Bilingual Education (MTbBE) rollout
- Brazil Ministry of Education — Portaria MEC nº 539 — Política Nacional de Educação Escolar Indígena nos Territórios Etnoeducacionais
- Singapore Ministry of Education — What Education in Singapore Looks Like — bilingual advantage
Green Map creates custom-edited map images using open geographic data sources such as geoBoundaries, Natural Earth, OpenStreetMap, and government open data. These maps are edited visual materials, not raw data files, and are provided for education, documents, presentations, and graphic reference.