Part I — Situation overview

At the joint press conference of the Ministry of Education and Children’s Affairs and the Education Authority on 8 September 2026, the Hungarian results of the 2025 PISA assessment (Programme for International Student Assessment) were made public. The study, repeated every three years and operated by the OECD (Organisation for Economic Co-operation and Development), measures the reading, mathematical and scientific competences of 15-year-olds. The Hungarian data show deterioration in all three areas, and in two of them the weakest result since the assessments began. The average score in reading fell by 21 points, to 452 points; that in mathematics by 14 points, to 459 points, the worst value since 2003; the science average declined from 486 to 480 points. Kristóf Velkey, head of the Public Education Analysis Department of the Education Authority, highlighted, to convey the order of magnitude, that one school year theoretically adds roughly 20 points to a pupil’s knowledge. The 2025 assessment also examined computer-based problem solving for the first time: Hungarian pupils achieved an average of 495 points here, significantly below the OECD average. On the same day it became known that Sándor Brassói, president of the Education Authority, had resigned on 1 September with effect from 1 October — according to his statement for human reasons, independently of any political or personal conflict.

The deterioration has long-standing antecedents, and this is what gives the topic its real stake. The education researcher Péter Radó, in a Facebook post, drew up the balance sheet against the “last year of peace”, 2009: minus 42 points in reading, 31 in mathematics and 23 in science. He highlighted separately that the proportion of failing pupils has risen sharply: in 2022 a quarter of pupils, and in 2025 already a third, proved to be functionally illiterate. This category denotes those who, although able to read, are not able to understand and use what they have read reliably. Judit Lannert, minister of education and children’s affairs, placed the data in a global context at the press conference: “there is a global crisis in education”, and in her view the Hungarian results “fit neatly” into the international trend, always a little below the average. Radó does not regard this argument as exculpatory: in his view it would have been precisely the task of education governance to strengthen the resilience of the system against the spread of digital device use and artificial intelligence, and this evidently did not happen by 2025. The minister announced several interventions as well: a review of the system of talent development, a new National Core Curriculum within a year and a half, reading promotion campaigns, a programme aimed at involving parents and a strategy under preparation on the use of artificial intelligence in schools.

MIAK’s reading: the fifteen-year-olds now measured have been at school roughly since 2016, so the deterioration arose in the period for whose education governance decisions the present ministerial leadership is not responsible. This fact, however, does not exempt but redirects. The political reflex leads to the same place on both sides: the petition demanding the minister’s dismissal and the “global crisis” explanation alike avoid the only substantive question — in which groups of pupils, in which regions and in which skills the results collapsed, and what intervention has a proven effect. PISA is the only internationally comparable feedback on the state of Hungarian public education that is independent of political cycles — the worst outcome is if this feedback turns into a question of personnel.

Part II — Foundations in the literature

Three authors provide the frame in which the present data can be interpreted. Andreas Schleicher, the OECD’s director for education and the creator of the PISA programme, refutes with data in his volume World Class (2018) the idea that performance is a direct function of spending: above a certain threshold no correlation can be shown between expenditure per pupil and the average result — his example is precisely the comparison of Hungary and Luxembourg. John Hattie, the New Zealand educationalist, synthesised the results of more than 800 meta-analyses. In his work Visible Learning (2008) he introduces the yardstick without which the education policy debate cannot be settled: the threshold value of 0.40 for effect size. This is needed because an effect above zero in itself proves nothing — in education practically every intervention improves things somewhat. And the World Bank’s report World Development Report 2018 points to the structural reason why deterioration can remain invisible for a long time: education systems routinely report on enrolment and on spending, not on learning — and what is not measured does not appear on politicians’ agendas either. The detailed treatment of the literature — author by author, with quotations — can be found in section 6.4 Literature in detail.

Part III — MIAK’s concrete proposal

MIAK proposes three measurable measures.

3.1 Making the PISA microdata accessible to researchers (within 90 days)

The strongest legitimation of programme point O3 — data-driven education development — is precisely the present figure. MIAK proposes that within ninety days the Education Authority make accessible on researcher request the pupil-level data file of the Hungarian PISA assessment, rendered unsuitable for personal identification, together with the associated background variables: settlement type, school type, the index describing the family’s socio-economic situation, and the pupil’s earlier school career. The name of the institution need not be published — for examining segregation and territorial effects, settlement type and region are sufficient. Let the format of the data publication follow the logic of the D2 open data programme: a machine-readable file, with a documented list of variables and a clear access procedure. Without this, the debate of the next twelve months will run out where the present one does: the participants will argue about the average score, while the substantive question is the dispersion — where the 21-point fall in reading arose. The central claim of the World Bank report (see 6.4.3) is precisely that without measurement the problem remains not only unsolved but invisible as well.

3.2 A ranking by effect size for the planned interventions (by the start of the public debate on the new National Core Curriculum)

The ministry announced five interventions: a review of talent development, a new core curriculum, reading promotion campaigns, a parental involvement programme and a strategy on the use of artificial intelligence in schools. MIAK’s proposal is that none of these should start until a one-page annex is attached to each of them answering: what international evidence is available on the effect of the intervention, how large the measured effect size is, and against what success will be measured. Hattie’s threshold of 0.40 (see 6.4.2) is not scientific fussiness here but a budgetary question: the programme points O6 on educating for thinking and O8 on performance-based teacher motivation designate precisely those types of intervention where the evidence base is strongest and which are suited to addressing the collapse in reading directly. Publishing the ranking is also a protection for the ministry: if the order is public and justified, the professional debate will be about the evidence, not about the minister’s sentences.

3.3 Separate presentation of territorial and social dispersion in the public analysis (within 12 months)

The most important property of Hungarian PISA results has for decades been not the average but the size of the difference between schools. MIAK proposes that the Education Authority present, in a separate chapter of the Hungarian country report published within twelve months, how the 21-point fall in reading is distributed among settlement types, regions and groups by family background, and in which groups the rise in the proportion of failing pupils from a quarter to a third occurred. Combining this breakdown with the data of the TE3 segregated-area map will show whether the deterioration is an even sinking or whether the average is being pulled down by the falling behind of a few groups of pupils — the two cases require entirely different interventions. The TE4 rural public service minimum and the FO8 human capital investment framework are linked to this breakdown: today’s deficit in basic skills will appear in ten to fifteen years as a labour market cost, and remedying it in adulthood is substantially more expensive than prevention at school.

The three proposals are bound together by the same principle: the quality of the education policy debate is decided not by the number of announced programmes but by whether there is anything to measure them against. The microdata say where the trouble arose. The effect size ranking says which intervention has a chance. The dispersion breakdown says with whom one has to begin. None of them restricts the ministry in what reform it chooses — only in making its choice impossible to judge afterwards.

Part IV — Expected effects and risks

Dimension Expected effect Risk
Education Researcher access to the microdata will within a few months bring independent analyses that make interventions more targeted; the effect size ranking steers resources towards directions that demonstrably work Data publicity produces bad news in the short run, which creates a political temptation to postpone; ranking may slow the timetable of the curriculum transformation
Society The territorial and social breakdown makes visible that behind the fall in the average there stands not an even sinking but probably the falling behind of groups — this improves the targeting of catch-up resources The published breakdown can also be used to stigmatise schools and districts if identification at institutional level does not remain excluded
Economy and labour market Following the proportion of functional illiteracy gives a ten- to fifteen-year forecast of labour market adaptability, which underpins the planning of training resources The effect is delayed: the result of today’s intervention will first be visible in the next PISA cycle, in 2028, a time span longer than political patience

The main tension of the package runs between speed and soundness. After data like these, enormous pressure weighs on the decision-maker to act at once, and the most spectacular step — curricular transformation, institutional reorganisation — is the one that can be announced fastest. At the same time the lesson of Schleicher and Hattie is precisely that spectacle and effect do not coincide: the effect size of system-level reorganisations is typically low, while their cost is high, and implementation ties up the system’s attention for years. The proposal tips towards the risk side if the publicity of the microdata fails to happen, and the new core curriculum is prepared without our knowing in which groups of pupils the deterioration arose. The second question for consideration is that of publicity at institutional level: MIAK deliberately does not ask for the publication of school names. The steering force of ranking is real, but stigmatising low-performing schools would hit disadvantaged pupils hardest — a breakdown by settlement type and region is sufficient for the policy purpose.

Part V — Measurability and summary

5.1 What is worth following? (suggested KPIs)

MIAK proposes four performance indicators (KPIs, in full: Key Performance Indicators) from which it will be visible in 12 and 36 months whether the response has been substantive:

  • Researcher access to the PISA microdata: the suggested target is that within ninety days a documented access procedure should exist, and within twelve months at least five independent research analyses should be produced from the file.
  • The proportion of failing pupils in reading: starting from the 2025 value of a third, this is the most important indicator, because the movement of the average score may conceal this group falling behind. It is worth following up to the 2028 assessment.
  • The proportion of planned interventions with an evidence annex attached: the suggested target is that by the start of the public debate on the new core curriculum all five announced programmes should have a public annex that also states an effect size.
  • The share of between-school performance differences within total dispersion: this indicator says whether the selectivity of the Hungarian system is falling. It is worth following in combination with the TE3 segregated-area data.

5.2 Summary

MIAK’s request in a single sentence: before anything is transformed, let them make visible where the deterioration arose, and show which intervention has a proven effect. Concretely: researcher access to the microdata within ninety days, an evidence annex for all five announced programmes by the start of the debate on the core curriculum, and a separate territorial-social breakdown in the Hungarian country report within twelve months. None of these requests is about the ministry slowing down — they are about our being able to know what it achieved.

Two MIAK foundational values are in play here. One is data-drivenness: a professional judgment on a 21-point fall in reading is impossible today because we see nothing of it beyond the average — the microdata are not researcher convenience but the only way in which the debate becomes a question of fact. The other is accountability: the measured deterioration arose in earlier government cycles, and MIAK records this even when the argument now protects the ministry in office. The yardstick remains the same at the next assessment too — but for the 2028 figure it will already be the decision-maker who chooses today among the interventions who is responsible.

Part VI — Reasoning and further sources

6.1 The framing of the press by spectrum

The economic band carried the topic in the greatest detail, and its framing was twofold. Portfolio worked up the data in two separate articles: in one it published the content of the ministry press conference, with the ministerial leadership’s intervention plans and the details of the assessment — among them that 29 per cent of Hungarian pupils reported noise and disorder during lessons, which is below the OECD average of 31 per cent, and that 5 per cent of pupils had experienced online harassment in the past year, and that these performed on average 21 points worse in science. The other article published exclusively the assessment of the education researcher Péter Radó, who called the deterioration a catastrophe comparable to the 2015 collapse, and the responsibility of education governance unambiguous. This duality is in itself informative: the same paper published the governmental and the critical reading on the same day, separately.

The liberal-left and public affairs band made the validity of the explanation the axis of its framing. HVG put the historic low point in the headline, and made the minister’s sentence — “education is in a global crisis, and Hungary fits neatly into this” — the backbone of the introduction, that is, it did not refute it but set it against the measured data. 24.hu carried the same sentence at headline level, and handled the resignation news in a separate article, together with Sándor Brassói’s full justification. 444.hu published the statement of resignation, supplemented with the context that a mass departure from the authority had begun earlier. The common element of the framing is that the papers treated the personnel news and the assessment data separately — none of them asserted a connection between the two as a fact.

The conservative band placed at the centre not the PISA data but the person and the sentences of the minister. Mandiner reported on the KDNP’s petition, in which the parliamentary group leader Bence Rétvári called the ministerial leadership’s statements about teachers — among them “I would replace all the head teachers” and the sentences questioning teachers’ independent thinking — unacceptable. The same paper also published the position of Erzsébet Nagy, president of the Democratic Trade Union of Teachers, who rejected both the closure of small schools and the mass replacement of head teachers, while indicating: it is not right to judge the minister’s present thinking on the basis of a single earlier sentence taken out of context. This last consideration is substantive by MIAK’s yardstick: the fate of small schools indeed cannot be decided on headcount alone, because the decision touches the TE4 rural public service minimum — in many small settlements the school is the last remaining public institution. MIAK’s answer is that it is precisely for this reason that the territorial breakdown is needed: decisions on mergers must be preceded not by a debate of principle but by a settlement-by-settlement impact assessment. At the same time the personnel debate does not substitute for the policy answer: the measured deterioration persists irrespective of the outcome of the dismissal petition.

6.2 Facts and data

Data Value Source
The publication of the results 8 September 2026, joint press conference of the ministry and the authority Portfolio, 8 September 2026
Reading average score 452 points (−21 points against the previous assessment) Portfolio, 8 September 2026
Mathematics average score 459 points (−14 points; the worst since 2003) Portfolio, 8 September 2026
Science average score 480 points (from 486) Portfolio, 8 September 2026
Computer-based problem solving (measured for the first time) 495 points, significantly below the OECD average Portfolio (Péter Radó’s analysis), 8 September 2026
Change against 2009 reading −42, mathematics −31, science −23 points Portfolio (Péter Radó’s analysis), 8 September 2026
The estimated added value of one school year about 20 points Kristóf Velkey, Education Authority / Portfolio, 8 September 2026
The proportion of failing pupils 2022: a quarter of pupils → 2025: a third Portfolio (Péter Radó’s analysis), 8 September 2026
The proportion of pupils reporting noise during lessons 29% (OECD average: 31%) Portfolio, 8 September 2026
The proportion of pupils suffering online harassment 5%; those concerned achieved on average a 21-point weaker science result Portfolio, 8 September 2026
The proportion of pupils using artificial intelligence to do the task for them 29% Portfolio, 8 September 2026
The resignation of the president of the Education Authority submitted on 1 September 2026, effective 1 October 2026 24.hu / 444.hu, 8 September 2026
The announced time horizon of the new National Core Curriculum 1–1.5 years Portfolio, 8 September 2026

Two data points require a separate note. Péter Radó’s analysis classifies the 6-point decline in science as not significant in the statistical sense, so there it is more justified to speak of stagnation or slight deterioration — the falls in reading and mathematics, by contrast, are substantial. And the labels “failing pupil” and “functionally illiterate” cover the same measurement category: those who in understanding what they read do not reach the level needed for independent use; the table follows the measurement category, not the colloquial label.

6.3 Policy dimensions

  • Education (programme points) — the measurement system and the evidence base of interventions: O3 (data-driven education development), O6 (educating for thinking), O8 (performance-based teacher motivation) and O2 (digital preparation of teachers) together provide the skeleton of the proposal;
  • Territorial inequality and rural policy (programme points) — the dispersion behind the average: TE3 (segregated-area map and catch-up programme) and TE4 (rural public service minimum) provide the two data systems with which the PISA breakdown can be combined;
  • Employment policy (programme points) — the delayed economic consequence: FO8 (human capital investment framework) shows how much more expensive remedying in adulthood is than prevention at school;
  • Digitalisation and AI regulation (programme points) — the form of data publication: D2 (open data programme) provides the technical frame of microdata access.

6.4 Literature in detail

6.4.1 Andreas Schleicher: World Class

The second chapter of Schleicher’s volume sets education policy commonplaces against the PISA data. The part on spending is the most important from the point of view of the topic: the author shows that a strong correlation between expenditure per pupil and results exists only up to a certain threshold — roughly 50 thousand dollars in total between the ages of 6 and 15 — while above this no connection can be demonstrated. The example is precisely the Hungarian figure:

“Fifteen-year-old students in Hungary, on whom 47 thousand dollars are spent between the ages of 6 and 15, perform at the same level as students in Luxembourg, on whom more than 187 thousand dollars are spent, even after taking account of differences in purchasing power parity.”

From this Schleicher draws the conclusion that success depends not on how much money is spent but on how. In the same chapter he also refutes the idea that a smaller class size in itself would bring better results: the best-performing systems, if they have to choose, invest in the quality of teachers, not in reducing class size.

In the Hungarian situation this means two things. On the one hand, the present deterioration cannot be explained by a simple lack of resources, and neither can it be resolved by additional resources alone — which does not exonerate the funding, but only shows that the question of how comes first. On the other hand, of the announced interventions the most promising are precisely those that affect the quality of teachers’ work, and not the structure.

📖 Source: Andreas Schleicher: World Class

6.4.2 John Hattie: Visible Learning

Hattie introduces his synthesis of more than 800 meta-analyses with the recognition that in education the question “does it work” is badly posed, because measured against zero practically every intervention improves things somewhat. He therefore proposes a new reference point:

“To set the bar at zero is absurd. […] Let us set the bar at an effect size of 0.40. The average effect size is 0.40, and this summarises the typical magnitude of all possible effects in education.”

The author calls this the “hinge-point”, and adds that it is not a magic number but the threshold above which the result means a change noticeable in reality as well. He brings homework as an illustrative example: its average effect size is 0.29, that is, measured against zero it “works”, but measured against the typical effect it is a weaker than average intervention — many other steps bring a greater result from the same outlay.

From the point of view of the Hungarian education policy debate this is the yardstick that is missing most. Of each of the five interventions now announced it can be said that it “improves the situation” — the question is which reaches the threshold and which stays below it. MIAK’s evidence annex proposal asks precisely for this ranking, before the reorganisation of the system, not afterwards.

📖 Source: John Hattie: Visible Learning

6.4.3 World Bank: World Development Report 2018

The fourth chapter of the report describes the absence of measurement not as a technical but as a political problem. The central claim is that education systems routinely report on enrolment, not on learning — and this has a consequence for the agenda:

“Because learning is absent from official education management data, it is also absent from the agenda of politicians and officials.”

The report adds that the absence of learning data allows governments to ignore or conceal the poor quality of education, especially in the case of disadvantaged groups, and that parents cannot hold to account what they do not see. It highlights separately that measurement is not the same as international tests: it is a toolkit ranging from classroom diagnostic assessments to national examinations, whose elements do not replace but complement one another.

The Hungarian case illustrates this mechanism. PISA was able to signal the deterioration because it is international and repeated every three years — the question is why the domestic measurement and feedback system did not signal the same thing earlier. The proposal on the publicity of microdata targets precisely this absence: it does not ask for a further assessment, but for the results of the assessment that already exists to become analysable.

📖 Source: World Bank: World Development Report 2018 — Learning to Realize Education’s Promise

6.5 International comparison

The most instructive parallel is the German “PISA shock”. The World Bank report cites it as an example too: in Germany the first PISA assessment in 2000 brought a weaker than expected result, especially among the children of poorer families and those with a migrant background, and this led to the building of a targeted support system for disadvantaged pupils. Two elements of the German response deserve attention. One is that the reaction was directed not at the average but at the dispersion — the interventions were targeted where the falling behind was measurable. The other is that, according to Schleicher’s analysis, PISA was able to identify the most serious structural feature of the German system, the choice of school type at the age of ten, as a problem because the data became analysable in combination with family background: it turned out that selection largely reproduced the existing social structure.

This second point bears directly on the Hungarian situation, where early selection and the large performance differences between schools have for decades been the most characteristic feature of the system. The lesson is not that the German package of measures can be taken over — the systems differ — but that the targeting of the intervention depended on data combined with background variables. In Hungary this combination is today not available to independent research, and because of this the next reform will have to be planned just as blindly as the previous one.

The third international lesson concerns the timing of measurement systems. In the Brazilian practice presented in the report, alongside a biennial national assessment there are feedback assessments running every two months and tied to learning stages, whose aim is not evaluation but support for teacher intervention. International data arriving every three years is, by comparison, too infrequent and too late to be suitable for correction during the school year — this argument speaks for strengthening the domestic measurement component of programme point O3.

Education

  • O2 — Digital preparation of teachers
  • O3 — Data-driven education development
  • O6 — Educating for thinking in the age of the division of labour
  • O8 — Performance-based teacher motivation

Territorial inequality and rural policy

  • TE3 — Segregated-area map and catch-up programme
  • TE4 — Rural public service minimum

Employment policy

  • FO8 — Human capital investment framework

Digitalisation and AI regulation

  • D2 — Open data programme

Suggested new programme point: An evidence annex for education interventions — for the Education area: every major education policy measure should have a public, one-page annex on the international evidence base, the expected effect size and the way success will be measured.

6.7 List of sources

Press sources (MIAK press monitor, 9 September 2026 — topic 2):

Knowledge base references (specialist books):

  • 📖 Andreas Schleicher: World Class
  • 📖 John Hattie: Visible Learning
  • 📖 World Bank: World Development Report 2018 — Learning to Realize Education’s Promise

MIAK internal materials:

  • MIAK policy area: Education (programme points; programme point ID: O2, O3, O6, O8)
  • MIAK policy area: Territorial inequality and rural policy (programme points; programme point ID: TE3, TE4)
  • MIAK policy area: Employment policy (programme points; programme point ID: FO8)
  • MIAK policy area: Digitalisation and AI regulation (programme points; programme point ID: D2)
  • MIAK press monitor, 9 September 2026 — topic 2, score: 90/100

Supplementary public data sources:

  • OECD PISA country profiles — the Hungarian time series in international comparison
  • The National Competence Assessment — the domestic measurement time series between PISA cycles

Generation metadata