School: what do the systems at the top of PISA do?
East Asia and Estonia have the highest scores in the PISA survey, with few pupils far behind. France has been slipping for twenty years. What these systems have in common, what has really been evaluated, and what does not set them apart.
Scope: Primary, lower and upper secondary: the 15-year-olds tested by PISA 2025 (OECD), higher education left out; spending per pupil from primary to upper secondary (Education at a Glance, OECD), latest year published for each country.
Key points
- The leading systems above all have few pupils far behind. Singapore (563 points), Japan (525 points), South Korea (522 points) and Estonia (508 points) are well above the OECD average (463 points). Their lead comes from the bottom as much as the top: 16.1% of pupils far behind in Japan and 17.2% in Estonia, against 35.7% in France.
- Keeping all pupils together longer lifts the weakest (evidence level B). Estonia, Japan and Korea do not split pupils before 15 or 16. In Poland, one more year of common schooling in 1999 was followed by a sharp rise in reading (479 points in 2000, 518 points in 2012), and studies attribute part of that gain to the reform.
- The best-demonstrated measures concern the classroom, not structures (level A). Small-group tutoring during the school day has large effects in dozens of controlled experiments; Singapore’s maths method, tested in England, gives a real but modest gain. Teacher quality matters a great deal (level B).
- Beyond a certain level, it is not about money. A pupil costs $15,798 a year in the United States and $13,722 in France, against $10,827 in Estonia. A score at least as high as France’s is observed elsewhere for 71% less spending. A correlation, not proof: within a country, money well spent improves results.
How to read these figures
PISA tests 15-year-olds every three years in maths, reading and science, whatever their grade. Scores are on a common scale; trends are read on the average of the OECD countries comparable in every round, which differs slightly from the average of all members. China is represented only by four rich provinces (Beijing, Shanghai, Jiangsu, Zhejiang), not by the whole country. A pupil “far behind” is below Level 2, the minimum the OECD considers needed to use maths in everyday life.
Each measure comes with a level of evidence: A for controlled experiments (random assignment), B for quasi-experimental studies (a reform compared with a control group), C for comparisons of data, D for a plausible, untested hypothesis. Charts across countries are correlations: culture, private lessons outside school, wealth and immigration matter too. They raise questions; they do not point to a model to copy.
1. Who leads, and who is slipping
In 2025 the highest maths scores are those of four Chinese provinces (612 points), Singapore (563 points), Macao (549 points), Chinese Taipei (546 points), Japan and South Korea. Estonia (508 points) is the first European country, 50 points ahead of France.
The average of the comparable OECD countries has lost 33 points since 2003 – 2025. France has lost 53 since 2003. Finland, long held up as a model, has fallen 80 points from its peak in 2006 and is now below Poland: a ranking is never secured.
Mean maths score in PISA, 15-year-olds
The OECD average is that of the countries with comparable data in every round. 2022 follows the school closures of the pandemic.
- OECD average (comparable countries)
- France
- Finland
- Poland
- Estonia
- South Korea
- Japan
Source:OECD
Show the chart data
| Period | Japan | South Korea | Estonia | Poland | Finland | France | OECD average (comparable countries) |
|---|---|---|---|---|---|---|---|
| 2003 | 534 points | 542 points | – | 490 points | 544 points | 511 points | 502 points |
| 2006 | 523 points | 548 points | 515 points | 495 points | 548 points | 496 points | 501 points |
| 2009 | 529 points | 546 points | 512 points | 495 points | 541 points | 497 points | 502 points |
| 2012 | 536 points | 554 points | 521 points | 518 points | 519 points | 495 points | 499 points |
| 2015 | 532 points | 524 points | 520 points | 505 points | 511 points | 493 points | 496 points |
| 2018 | 527 points | 526 points | 523 points | 516 points | 507 points | 495 points | 496 points |
| 2022 | 536 points | 527 points | 510 points | 489 points | 484 points | 474 points | 480 points |
| 2025 | 525 points | 522 points | 508 points | 484 points | 469 points | 458 points | 469 points |
2. A high level and few pupils left behind
The leading systems do not owe their rank to an elite: they have very few pupils below Level 2 (3.6% in the Chinese provinces, 11.2% in Singapore, 16.1% in Japan, 17.2% in Estonia, against 35.7% in France and 34.9% in the OECD). They also have more top performers (37.4% in Singapore, 22.3% in Japan, 5.2% in France).
The gap linked to social background varies widely. It is 76 points in Estonia, 79 in Japan, 55 in Macao, against 95 in France and 112 in Germany. The chart below places the 86 participants: a high level and a moderate social gap can go together (Macao, Estonia, Japan), though not as a rule, since Singapore’s gap is 103 points.
Gap linked to social background and maths score, PISA 2025 participants
Horizontally: difference in maths score between the most and the least advantaged quarter of pupils (OECD index of economic, social and cultural status). Vertically: mean score. Regions tested alone are not shown, except the four Chinese provinces.
Source:OECD
Show the chart data
| Country | Social gap (points) | Maths score | Period |
|---|---|---|---|
| Kenya | 3 points | 326 points | 2025 |
| Cambodia | 20 points | 366 points | 2025 |
| Rwanda | 21 points | 312 points | 2025 |
| Jordan | 31 points | 388 points | 2025 |
| Kosovo | 39 points | 350 points | 2025 |
| Saudi Arabia | 41 points | 390 points | 2025 |
| Kyrgyzstan | 41 points | 364 points | 2025 |
| Indonesia | 43 points | 364 points | 2025 |
| Armenia | 43 points | 376 points | 2025 |
| Kazakhstan | 44 points | 414 points | 2025 |
| Thailand | 46 points | 407 points | 2025 |
| Palestinian Territories | 46 points | 354 points | 2025 |
| Georgia | 50 points | 416 points | 2025 |
| Philippines | 51 points | 371 points | 2025 |
| United Arab Emirates | 51 points | 453 points | 2025 |
| Dominican Republic | 53 points | 339 points | 2025 |
| Lebanon | 54 points | 377 points | 2025 |
| Macao (China) | 55 points | 549 points | 2025 |
| Guatemala | 55 points | 334 points | 2025 |
| Morocco | 55 points | 355 points | 2025 |
| Albania | 57 points | 433 points | 2025 |
| Mexico | 57 points | 388 points | 2025 |
| Qatar | 58 points | 416 points | 2025 |
| Denmark | 58 points | 471 points | 2025 |
| Latvia | 58 points | 461 points | 2025 |
| Iceland | 58 points | 450 points | 2025 |
| El Salvador | 59 points | 346 points | 2025 |
| Paraguay | 61 points | 334 points | 2025 |
| Montenegro | 62 points | 411 points | 2025 |
| Canada | 63 points | 485 points | 2025 |
| North Macedonia | 64 points | 380 points | 2025 |
| Lithuania | 67 points | 471 points | 2025 |
| Slovenia | 67 points | 460 points | 2025 |
| Hong Kong SAR China | 67 points | 522 points | 2025 |
| Ecuador | 68 points | 366 points | 2025 |
| Malta | 68 points | 439 points | 2025 |
| Chile | 70 points | 403 points | 2025 |
| Cyprus | 70 points | 412 points | 2025 |
| Finland | 71 points | 469 points | 2025 |
| Ireland | 73 points | 480 points | 2025 |
| Spain | 73 points | 457 points | 2025 |
| Azerbaijan | 73 points | 422 points | 2025 |
| Peru | 75 points | 382 points | 2025 |
| Brunei | 76 points | 435 points | 2025 |
| Estonia | 76 points | 508 points | 2025 |
| Greece | 77 points | 424 points | 2025 |
| Colombia | 78 points | 381 points | 2025 |
| Mauritius | 78 points | 407 points | 2025 |
| Norway | 78 points | 452 points | 2025 |
| Moldova | 78 points | 410 points | 2025 |
| Serbia | 79 points | 417 points | 2025 |
| Japan | 79 points | 525 points | 2025 |
| Italy | 79 points | 468 points | 2025 |
| Argentina | 79 points | 367 points | 2025 |
| Mongolia | 80 points | 419 points | 2025 |
| Sweden | 80 points | 464 points | 2025 |
| Austria | 81 points | 477 points | 2025 |
| Poland | 82 points | 484 points | 2025 |
| Australia | 82 points | 478 points | 2025 |
| Israel | 83 points | 433 points | 2025 |
| Malaysia | 83 points | 397 points | 2025 |
| Beijing-Shanghai-Jiangsu-Zhejiang (China) | 84 points | 612 points | 2025 |
| New Zealand | 86 points | 480 points | 2025 |
| United Kingdom | 87 points | 488 points | 2025 |
| Croatia | 88 points | 455 points | 2025 |
| Brazil | 90 points | 377 points | 2025 |
| South Korea | 90 points | 522 points | 2025 |
| Bulgaria | 91 points | 405 points | 2025 |
| Vietnam | 93 points | 443 points | 2025 |
| United States | 93 points | 463 points | 2025 |
| Uruguay | 93 points | 405 points | 2025 |
| Chinese Taipei | 94 points | 546 points | 2025 |
| France | 95 points | 458 points | 2025 |
| Türkiye | 96 points | 462 points | 2025 |
| Belgium | 97 points | 480 points | 2025 |
| Switzerland | 98 points | 499 points | 2025 |
| Portugal | 100 points | 460 points | 2025 |
| Czechia | 100 points | 477 points | 2025 |
| Netherlands | 102 points | 483 points | 2025 |
| Singapore | 103 points | 563 points | 2025 |
| Hungary | 104 points | 460 points | 2025 |
| Luxembourg | 108 points | 458 points | 2025 |
| Germany | 112 points | 464 points | 2025 |
| Slovakia | 116 points | 469 points | 2025 |
| Romania | 135 points | 419 points | 2025 |
| Zambia | 143 points | 314 points | 2025 |
3. Keeping all pupils together longer
Estonia schools every pupil in a single basic school for nine years; Japan and South Korea do not split them before upper secondary. Germany, by contrast, sends most children into different tracks at around 10, and its social gap is among the widest in the previous chart.
Poland is the best-studied case. In 1999 it created a single lower secondary school that delayed vocational tracking by one year. Its reading results rose from 479 points in 2000 to 518 points in 2012, above the OECD average (500 points in 2000). In 2017 the country went back to tracking at 15; its reading score is 482 points in 2025, a fall the pandemic makes impossible to attribute to the reform alone.
Common schooling is not a universal rule of the leading systems: Singapore long split its pupils into tracks from lower secondary and has only been replacing them since 2024 with subject-by-subject levels.
Mean reading score in PISA, Poland and comparisons
Poland’s 1999 reform affected pupils tested from 2003 on; the 2017 reform, pupils tested from 2022 on.
- OECD average (comparable countries)
- Germany
- France
- Finland
- Poland
Source:OECD
Show the chart data
| Period | Poland | Finland | France | Germany | OECD average (comparable countries) |
|---|---|---|---|---|---|
| 2000 | 479 points | 547 points | 505 points | 484 points | 500 points |
| 2003 | 497 points | 544 points | 496 points | 491 points | 497 points |
| 2006 | 508 points | 547 points | 488 points | 495 points | 496 points |
| 2009 | 501 points | 536 points | 496 points | 497 points | 499 points |
| 2012 | 518 points | 524 points | 506 points | 508 points | 501 points |
| 2015 | 506 points | 526 points | 499 points | 509 points | 497 points |
| 2018 | 512 points | 520 points | 493 points | 498 points | 493 points |
| 2022 | 489 points | 490 points | 474 points | 480 points | 482 points |
| 2025 | 482 points | 474 points | 456 points | 465 points | 466 points |
Sorting pupils later: what the research says
Two kinds of studies agree: delaying the split of pupils into tracks narrows the gaps between them, and Poland’s 1999 reform, the best studied, also raised the average. These are quasi-experimental studies, not controlled experiments: the result is solid, its size depends on the country.
- BJakubowski, Patrinos, Porta and Wiśniewski (2016), Education EconomicsMethod: Poland’s 1999 reform (one more year of common schooling), pupils before and after compared by matching and difference-in-differences
Delaying vocational tracking and spending more time on basic skills had a positive, significant effect on pupils’ results, above all for those who would have gone into vocational tracks.
- BHanushek and Woessmann (2006), Economic JournalMethod: international comparison of how gaps change between primary and secondary school, by age of first tracking
Early tracking increases inequality of results; its effect on the average is less clear, if anything negative.
The scale
- A: causal effect shown by several studies with a rigorous design (experiment, reform compared with a control group), in several countries.
- B: causal effect shown by at least one rigorous study, or several quasi-experimental studies that agree.
- C: strong correlations and a plausible mechanism, no causal proof.
- D: descriptive observation or expert opinion only.
4. Trained teachers and a coherent curriculum
The leading Asian systems take care over entry into the profession and teachers’ teamwork. Singapore trains all its teachers in a single national institute; in Japan and Shanghai, teachers prepare and observe lessons together, then improve them (“lesson study”). Their maths curricula move slowly, on few concepts at a time, until the whole class has mastered them.
These choices are consistent with what research finds on the effect of teachers. But only the maths method has been tested outside Asia, with a modest gain: copying one practice without the rest of the system delivers a small part of it.
Teachers and methods: what the research says
That teacher quality matters a great deal is established by large quasi-experimental studies. Whether a method from Asia works elsewhere has been tested in controlled experiments: the effect is positive but modest. The selective recruitment practised in Asia has never been evaluated experimentally.
- BChetty, Friedman and Rockoff (2014), American Economic ReviewMethod: one million US pupils followed into adulthood, each teacher’s effect measured by the progress of their classes
Pupils who had a teacher whose classes progress a lot go to college more often and earn more as adults.
- AJerrim and Vignoles (2016), Economics of Education ReviewMethod: two controlled experiments in English primary and secondary schools with the Singapore-inspired “mastery” maths method
The method slightly improves maths results at low cost: a real but modest gain.
The scale
- A: causal effect shown by several studies with a rigorous design (experiment, reform compared with a control group), in several countries.
- B: causal effect shown by at least one rigorous study, or several quasi-experimental studies that agree.
- C: strong correlations and a plausible mechanism, no causal proof.
- D: descriptive observation or expert opinion only.
5. Catching struggling pupils early
The clearest common feature of the leading systems is the small number of pupils far behind. The best-demonstrated intervention to get there is tutoring: a few pupils with a trained adult, regularly, during school hours.
Part of the Asian lead is, however, won outside school: in Japan, South Korea, Singapore and China, many families pay for private evening lessons. Scores measure what pupils know, not only what school taught them, and this pressure has a cost for their well-being.
Helping struggling pupils early: what the research says
This is one of the best-demonstrated school interventions: dozens of controlled experiments, pooled in a meta-analysis, find large effects. They come mostly from the United States and the United Kingdom, not from Asian countries.
- ANickow, Oreopoulos and Quan (2020), NBERMethod: meta-analysis of nearly a hundred controlled experiments of tutoring, from pre-school to upper secondary
Tutoring has large effects on learning, larger when it is given by teachers or trained assistants, during the school day, and early for reading.
The scale
- A: causal effect shown by several studies with a rigorous design (experiment, reform compared with a control group), in several countries.
- B: causal effect shown by at least one rigorous study, or several quasi-experimental studies that agree.
- C: strong correlations and a plausible mechanism, no causal proof.
- D: descriptive observation or expert opinion only.
6. What does not set the leading systems apart: money
A pupil from primary to upper secondary costs $15,798 a year in the United States, $13,722 in France, $11,863 in Japan, $22,486 in South Korea, $10,827 in Estonia and $10,423 in Poland. Of the 37 countries observed, 7 form the efficiency frontier, and a score at least as high as France’s is reached elsewhere for 71% less spending (38% for the United States).
Class size does not set them apart either: Japan’s lower secondary classes are among the largest in the OECD. That does not mean money and small classes are useless: the best studies show they help, above all the youngest and the poorest. They simply do not explain the gaps between countries.
Spending per pupil and maths score, OECD and partner countries
Annual spending of institutions per pupil, primary to upper secondary (Education at a Glance, OECD), purchasing-power-parity dollars, latest year published (2020 – 2023). PISA 2025 score. The dotted line is the efficiency frontier.
Source:OECD
Show the chart data
| Country | Spending per pupil (PPP $) | Maths score | Period | On the frontier |
|---|---|---|---|---|
| Mexico | $3,406 | 388 points | 2022 | yes |
| Türkiye | $4,032 | 462 points | 2022 | yes |
| Romania | $6,288 | 419 points | 2022 | |
| Chile | $6,602 | 403 points | 2022 | |
| Bulgaria | $7,624 | 405 points | 2022 | |
| Hungary | $7,900 | 460 points | 2022 | |
| Greece | $8,119 | 424 points | 2022 | |
| Latvia | $8,355 | 461 points | 2022 | |
| Slovakia | $9,784 | 469 points | 2022 | yes |
| Lithuania | $10,030 | 471 points | 2022 | yes |
| Poland | $10,423 | 484 points | 2022 | yes |
| Estonia | $10,827 | 508 points | 2022 | yes |
| New Zealand | $11,760 | 480 points | 2023 | |
| Czechia | $11,847 | 477 points | 2022 | |
| Japan | $11,863 | 525 points | 2022 | yes |
| Spain | $12,231 | 457 points | 2022 | |
| Israel | $12,372 | 433 points | 2022 | |
| Portugal | $12,586 | 460 points | 2022 | |
| Canada | $12,765 | 485 points | 2020 | |
| Slovenia | $13,005 | 460 points | 2022 | |
| Italy | $13,459 | 468 points | 2022 | |
| Finland | $13,508 | 469 points | 2022 | |
| France | $13,722 | 458 points | 2022 | |
| Ireland | $14,059 | 480 points | 2022 | |
| United Kingdom | $15,416 | 488 points | 2022 | |
| Sweden | $15,484 | 464 points | 2022 | |
| Australia | $15,611 | 478 points | 2022 | |
| United States | $15,798 | 463 points | 2021 | |
| Germany | $16,433 | 464 points | 2022 | |
| Denmark | $16,783 | 471 points | 2023 | |
| Belgium | $17,224 | 480 points | 2022 | |
| Netherlands | $17,287 | 483 points | 2022 | |
| Iceland | $18,287 | 450 points | 2022 | |
| Austria | $18,626 | 477 points | 2022 | |
| Norway | $20,407 | 452 points | 2022 | |
| South Korea | $22,486 | 522 points | 2022 | |
| Luxembourg | $29,238 | 458 points | 2022 |
- 71%Less spending for a score at least as high as France’scheapest country among those doing at least as well ($4,032 per pupil)
Source:OECD
- 38%Same calculation for the United Statesa score at least as high is observed elsewhere for a fraction of US spending
Source:OECD
Money and class size: what the research says
Comparisons between countries show no clear link between spending and results beyond a certain level. But within a country, experiments and well-studied reforms show that money well spent, and smaller classes for the youngest, improve results. It is not what sets the leading systems apart; that does not make it useless.
- AKrueger (1999), Quarterly Journal of Economics (Project STAR, Tennessee)Method: controlled experiment: about 11,600 pupils from kindergarten to third grade randomly assigned to small or regular classes
Pupils in small classes progress more in the first year and then a little each year, above all disadvantaged pupils.
- BJackson, Johnson and Persico (2016), Quarterly Journal of EconomicsMethod: school finance reforms in US states, cohorts exposed or not compared
A lasting rise in spending per pupil lengthens schooling and raises adult wages, above all for children from low-income families.
The scale
- A: causal effect shown by several studies with a rigorous design (experiment, reform compared with a control group), in several countries.
- B: causal effect shown by at least one rigorous study, or several quasi-experimental studies that agree.
- C: strong correlations and a plausible mechanism, no causal proof.
- D: descriptive observation or expert opinion only.
What we know, what we don’t
What the figures establish
- The systems at the top of PISA have very few pupils far behind; several of them also have a moderate social gap.
- Delaying tracking reduces inequality; in Poland, the 1999 reform also raised the average.
- Small-group tutoring during the school day and teacher quality have demonstrated effects.
What they don’t tell
- How much of the Asian lead comes from school rather than private lessons, culture or pupils’ working hours.
- Whether the practices of the leading systems would give the same results transplanted wholesale into another country.
- What explains the recent decline of Finland, France and the OECD average, beyond the school closures of the pandemic.
Sources
Series downloaded from their producers on Nov 15, 2026, the snapshot date. Only published periods are used: no projection.