A student can reach the answer and still be unable to show why it is right. Another can follow a passage but stop when asked to compare two expressions, identify textual evidence and describe the effect of a word. Japan’s 2026 National Assessment of Academic Ability did not show a country whose children know nothing. It showed a system in which many can execute learned procedures, yet encounter a persistent barrier when they must select, connect and communicate the reasons behind them.
The National Institute for Educational Policy Research released the first of three result stages on July 16. The tests ran from April 20 through May 29. Grade 6 students took Japanese and mathematics; Grade 9 students took Japanese, mathematics and English. The initial release supplied national averages, English IRT scores and bands, and preliminary questionnaire findings. Full national analysis is scheduled for August 3, followed by prefecture and designated-city analysis in the fall.
That timetable sets an important boundary around this July 18 report. It is too early to assign causes to family background, teaching practice, region or gender. But the questions, answer keys and breakdowns by curriculum domain, assessment objective and response format are public. Together they reveal a pattern: performance falls when students must compare representations, prove a claim, interpret data or construct an evidence-based explanation.
The national averages are not pass marks
7.9 of 13 items
9.1 of 16 items
9.6 of 15 items
9.2 of 16 items
It is misleading to convert these percentages into the pass-fail language of an ordinary classroom exam. The national assessment does not sample every textbook topic evenly. A small number of questions deliberately probes the curriculum’s harder aspirations: applying knowledge, selecting information, interpreting evidence, making a judgment and explaining it. There is no official pass line, and demanding questions are part of the design.
Year-to-year comparisons require equal care. From 2025 to 2026, the raw Grade 6 Japanese average appears to have moved from 67.0% to 61.1%, while Grade 9 mathematics appears to have risen from 48.8% to 57.4%. Yet the texts, conditions, number of items and difficulty change each year. A full census produces a very precise description of the students who took that year’s paper; it does not make different papers a common longitudinal scale. “Scores fell” and “scores recovered” are conclusions the raw rates cannot support by themselves.
The central 2026 finding is not that averages sat in the 50s. It is the distance between carrying out a familiar procedure and selecting evidence to explain why the answer works.
“Writing” and “constructed response” are not synonyms
Two official classifications are often collapsed in coverage of Japanese-language results. The first is the curriculum domain called “writing.” The second is the response format called “constructed response.” A reading question may require a student to write an explanation and therefore count as constructed response. A question about how writing is organized may use a selected response. The categories overlap, but they do not measure the same set of items.
| Japanese | Overall | Writing domain | Constructed response |
|---|---|---|---|
| Grade 6 | 61.1% | 45.8% | 55.1% |
| Grade 9 | 64.2% | 67.7% | 45.9% |
The Grade 9 difference—67.7% for the formal writing domain and 45.9% for constructed responses—is not a contradiction. Some constructed-response items assess reading or speaking-and-listening content. The broad concern is therefore more precise than “students cannot write essays.” Many students struggle with a compound task: select the right material, return to the exact wording of a passage, satisfy a specified perspective and length, and organize the answer in language another person can follow.
One Grade 9 item made the problem visible. Students compared two examples and explained both the concrete state conveyed by the mimetic word kippari—roughly, firmly or decisively—and the effect of that expression. Only 35.5% answered correctly; 19.5% left it blank. Knowing a dictionary gloss was insufficient. Students had to infer attitude from context, compare passages and turn that interpretation into a bounded explanation.
Grade 6 mathematics: moving among ratios, diagrams and language
The Grade 6 mathematics average was 56.6%. By content domain, number and calculation stood at 61.3%, geometry at 55.5% and measurement at 78.2%. Change and relationships fell to 41.4%, while data use stood at 48.8%. By assessment objective, knowledge and skills reached 60.3%, compared with 45.6% for thinking, judgment and expression. Selected response was 56.2%, short answer 59.9% and constructed response 48.7%.
Ratio tasks demand more than recalling a formula. One situation stated that 14 turtles represented 20% of a total and asked students to form an equation and find the whole. Another asked them to select two diagrams in which a red tape was 1.5 times as long as a white tape. Students had to identify the reference quantity—what counts as one—and move accurately among a picture, numbers, a relationship and an equation.
Questions about crowding and parcel-delivery conditions made the same demand in everyday clothing. A familiar context is not necessarily an easy one. Students must discard irrelevant information, align units, choose a comparison and check whether the result makes sense in the situation. A formula begins the work; an understanding of what the formula represents allows it to travel to a new context.
Grade 9 mathematics: the 63.3–39.6 divide
The sharpest internal gap appeared in Grade 9 mathematics. The overall rate was 57.4%. Knowledge and skills reached 63.3%; thinking, judgment and expression reached 39.6%. Selected response was 59.6% and short answer 64.6%, while constructed response was 39.6%. There was a gap of almost 24 percentage points between using mathematical knowledge and expressing the reasoning that makes its use defensible.
| Work the student had to perform | What made it difficult |
|---|---|
| Explain how to estimate the duration of a 1,000MB video from a graph | Read the proportional relationship and state which values are used in an order another reader can reproduce. |
| Prove that angles are equal using triangle congruence | Extract conditions from the diagram and connect premise, theorem and conclusion without skipping the logical bridge. |
| Describe a pattern in frequency polygons about sleep and late bedtimes | Compare distributions rather than isolated points and keep the claim within what the data support. |
Calling the result a generic writing problem would miss part of the diagnosis. Mathematical explanation combines conceptual understanding, visual reading, logical sequence, vocabulary, notation and communication. A student may calculate correctly but omit a necessary condition, give only the conclusion, infer causation from a graph that shows association, or cite congruence without completing the chain of proof.
Nor does an emphasis on reasoning mean reducing procedural practice. If basic operations consume all of a learner’s working memory, little remains for an explanation. The productive alternative is not calculation or thinking. It is fluent procedure followed by questions such as: Why is this operation appropriate? When would it fail? Can another representation show the same relationship? Is the answer plausible?
English placed all four skills on screen
The 2026 Grade 9 English assessment marked a different kind of turning point. For the first time, listening, reading, writing and speaking were all administered by computer. The national four-skill IRT mean was 499. The distribution placed 6.4% in Band 1, 27.1% in Band 2, 36.2% in Band 3, 24.7% in Band 4 and 5.7% in Band 5.
A score of 499 is not a 49.9% correct rate and it is not a pass mark. Item response theory models such properties as item difficulty to place performance from different forms on a common scale. Schools mainly receive scores for listening, reading and writing. The four-skill national estimate uses the sample that completed speaking on the same day, while speaking was otherwise administered across a window. Its method is fundamentally different from the paper Japanese and mathematics percentage-correct figures.
CBT can present sound, video, interactive displays and multiple forms, and it can return results faster. It also introduces device and network reliability, keyboard familiarity, audio conditions, accessibility, screen-reading effects and test security into the measurement. If writing and speaking are scored at national scale with automated assistance, scoring rules, human oversight, bias audits, appeals and personal-data safeguards become part of educational validity—not merely technical administration.
1956: postwar national testing began with samples
The history predates today’s system by half a century. The archive maintained by NIER shows that the 1956 national achievement survey sampled about 4.2% of Grade 6 and Grade 9 students for Japanese and mathematics. Its stated purpose was to understand achievement nationally and produce evidence for improving instruction, curriculum and educational conditions. Across the elementary, junior-high and high-school programs, other years also examined social studies, science, English, music and additional subjects.
The decisive change came in 1961. Japan administered Japanese, social studies, mathematics, science and English to nearly all second- and third-year junior-high students, continuing the near-census model through 1964. A census could reveal educational inequality and local conditions. It could also make schools and places directly comparable. Opposition focused on ranking, central control over education, teacher evaluation and teaching to the test. Confrontations and litigation became a major postwar education dispute, eventually reaching the Supreme Court’s Grand Bench in the 1976 Asahikawa test case.
The program returned to an approximately 20% sample in 1965 and 1966, after which that national series ended. Two questions from the era remain unresolved. How should a national government verify equal opportunity and minimum standards in compulsory education? And how can a diagnostic number be prevented from becoming a league table?
The PISA shock and the shorthand of “yutori”
The curriculum guidelines announced in 1998 and fully implemented from 2002 reduced and reorganized content under a five-day school week, created integrated-study time and promoted the broader idea of ikiru chikara—the capacity to live and learn. The program later became known as “yutori education.” Its stated purpose was not to abandon thought, but to create room for experience, inquiry and learning initiated by the student.
Then came the results published at the end of 2004. In PISA 2003, Japan placed sixth in mathematical literacy, after having ranked first in 2000, and 14th in reading, down from eighth. Scientific literacy remained second. Japan was still an internationally high performer, but the “PISA shock” intensified claims of declining achievement and attacks on yutori.
It is too simple to attribute a movement in one international cycle to a single curriculum. PISA samples 15-year-olds and asks how well they apply knowledge in unfamiliar real-life contexts. Participants, questions and social circumstances change. Yet the results affected policy. MEXT’s official account of the modern national test’s origins cites PISA and TIMSS 2003 and, crucially, a 2004 curriculum survey that found problems in Japanese constructed responses and junior-high mathematics. The two weaknesses prominent in 2026 were written into the rationale for the system that began in 2007.
The nationwide test returned in 2007
| Year | Turning point in the modern national assessment |
|---|---|
| 2007 | The current program began for Grade 6 and Grade 9 Japanese and mathematics. It was a full census through 2009. |
| 2010 and 2012 | Approximately 30% samples, supplemented by voluntary use. The planned 2011 survey was not administered as a national assessment after the Great East Japan Earthquake. |
| 2013 | A “detailed survey” again covered all eligible students and added studies of trends, family circumstances and policies such as class size. |
| 2014 onward | The census model returned. Results go to boards and schools; each student receives an individual report. |
| 2019 | English was added. Separate “Knowledge A” and “Application B” booklets gave way to questions integrating knowledge and application. |
| 2020 | The survey was canceled during COVID-19 school disruption. |
| 2025–26 | Grade 9 science, followed by all four Grade 9 English skills, moved into CBT and IRT reporting. |
The formal purposes are equal opportunity and standards in compulsory education, evaluation of policy, improvement of school instruction and a continuing cycle of review. National and regional findings are published; boards and schools receive their own results; students receive individual reports. It is not an entrance examination. It is a diagnostic instrument aimed simultaneously at national policy, local administration, classrooms and learners.
Full coverage nevertheless creates permanent tension. It can reveal small differences between schools and localities, which may support accountability. The same visibility can encourage score competition, excessive test preparation, exclusion of lower-performing students, concentration of personal data and administrative burden. The governance that keeps diagnosis from becoming rank-order punishment is as important as the quality of the test.
The curriculum’s three pillars meet the classroom
The curriculum guidelines revised from 2017 through 2019 organize competencies into three pillars: knowledge and skills; thinking, judgment and expression; and motivation to learn and humanity. Full implementation began in elementary schools in 2020, junior-high schools in 2021 and high schools progressively in 2022. The curriculum makes language a responsibility across subjects, strengthens statistics and problem-solving, and calls for active, interactive and deep learning.
The national test’s breakdowns are one way to ask whether that design has reached daily instruction. The 2026 preliminary questionnaire found that students reporting more self-directed engagement in solving tasks tended to have higher scores. Positive responses about challenge and curiosity also aligned with higher performance. These are correlations. The first release cannot tell whether engagement produces achievement, achievement produces confidence, a good classroom produces both, or family and other factors contribute.
The GIGA School program and pandemic accelerated one-device-per-student access. Devices can help learners share reasoning, compare representations, revise prose and manipulate data. They can also become machines for searching answers and submitting short reactions. Technology amplifies the design of learning; it does not supply the design.
PISA 2022 strength and domestic weakness can coexist
Is Japanese mathematics in collapse? An international measure requires a more balanced answer. In PISA 2022, Japan scored 536 in mathematical literacy, 516 in reading and 547 in science. That placed the country first, second and first among OECD members, and fifth, third and second among all participating systems. A set of difficult national items cannot turn those results into evidence of educational collapse.
At the same time, even students in high-performing systems miss demanding reasoning items. PISA samples 15-year-olds; the national assessment covers nearly all Grade 6 and Grade 9 students. Their purposes, ages, content and scales differ. A strong national average and a specific weakness in written reasoning can both be true. A high-performing system should be able to use diagnostic items to identify the next layer of improvement.
PISA 2022 itself adds nuance. Japanese students reported less confidence than the OECD average in solving real-life problems with mathematics and fewer experiences connecting mathematics to real-world events. International rank should not produce complacency; a low response rate on one domestic item should not produce fatalism. Each tells only part of the story.
What instruction can do—beyond “write more”
| Direction | A practical classroom move |
|---|---|
| Write short evidence statements every day | Use two- to four-sentence claim-evidence-reasoning responses in science, social studies and mathematics as well as Japanese. |
| Bridge speech to prose | Explain to a partner, annotate a diagram, then write individually. Locate the point where an oral idea fails to become text. |
| Move among representations | Show a ratio as a diagram, table, equation and sentence; ask which forms encode the same relationship. |
| Use wrong answers as curriculum | Discuss anonymized work: What part is valid? Which premise or connection is missing? |
| Share the criteria before the task | Let students use a short rubric for conditions, evidence, vocabulary and conclusion, then revise. |
| Teach vocabulary through effect | Replace a word such as kippari with near-synonyms and compare how attitude and reader impression change. |
| Protect procedural fluency | Retain focused practice, then ask why the equation fits, whether another method works and whether the answer is reasonable. |
Adding more constructed-response questions can merely measure the gap again. Teachers need time to model intermediate thinking, break down the architecture of a strong answer and let students revise after feedback. Expression should move from a one-shot product that receives a mark to a process of talk, drafting, checking and rewriting.
Teacher workload is a binding constraint. Serious reading of student explanations requires time and shared standards. Schools can use collaborative moderation, compact common rubrics, peer review during lessons and annotated examples so that scoring leads into the next instructional move. If AI assists drafting or scoring, humans must retain final judgment, audit bias, protect student data and explain what the system evaluated.
Equity: a blank answer does not measure ability alone
Constructed responses reveal richer thinking than selected responses, but they can also reveal unequal opportunities. Vocabulary at home, reading experience, the age at which a student began learning Japanese, disability, handwriting or keyboard difficulty, processing speed and test anxiety all enter the response. A learner may understand a mathematical idea while carrying an additional burden in producing a long Japanese explanation.
Fair support does not mean supplying the answer. It means reducing obstacles that are not the target of the measurement. Read-aloud or enlarged materials where appropriate, alternative input, time and quiet conditions, explicit teaching of academic Japanese for multilingual learners, and multiple forms of classroom assessment can help. As testing moves from paper to screen, mode effects and accommodations must be evaluated again.
A national average compresses approximately 1.8 million individuals into one number. Improvement begins by looking inside it: students who left an item blank, those who reached a correct intermediate step, and those who struggle with only one response mode. The August analysis should make distributions and error patterns at least as prominent as means.
All-subject CBT from 2027: comparability and new risk
MEXT revised its roadmap for computer-based national assessment in June 2026. From fiscal 2027, subjects are due to move into CBT, with item response theory intended to support stronger comparison over time as the calibrated scale matures. If common items anchor different forms, Japan can adjust for difficulty and follow change more defensibly than by comparing raw rates from different papers.
The potential is substantial. Today, a five-point movement may reflect students, questions or both. Multiple digital forms can cover more curriculum within the same testing time, reduce exposure of a single booklet and return more detailed information faster. A report can identify patterns in what a student can do instead of supplying one undifferentiated mark.
IRT is not magic. Paper-to-screen effects, device operation, outages, reduced public release of questions, algorithmic scoring and the construction of the scale must be explained. A national assessment is public educational infrastructure. Its technical specifications, uncertainty, validity, accessibility and data retention require scrutiny that schools and families—not only psychometricians and vendors—can understand.
What to watch on August 3 and in the fall
| Evidence to examine | Why it matters |
|---|---|
| Item-level error patterns | Separates missing knowledge, overlooked conditions, absent evidence and incomplete expression. |
| Distribution of blanks | May help distinguish not knowing from time, input barriers or disengagement. |
| Questionnaire relationships | Requires multiple-factor analysis rather than turning correlation into causation. |
| Prefecture and city variation | Common misconceptions and local patterns matter more than a rank table. |
| English CBT implementation | Tests effects of audio, input, connectivity, administration dates and the speaking sample. |
| Accommodations and missing data | Shows who could not participate fully and what support was used. |
| The 2027 CBT design | Needs detail on trend scales, public questions, scoring, privacy and disability access. |
For a school, the most useful result is not national position. It is the pattern of partial answers on particular tasks. Boards of education will do more good by protecting time for teachers to examine student work together and redesign the next unit than by rewarding schools for marginal movements in an average.
Conclusion: the achievement beyond a correct answer
Japan’s national testing history runs from the 1956 sample through the near-census tests and conflicts of the 1960s, anxiety around international studies and the program’s return in 2007. The declared purpose has remained recognizable: understand national standards and improve educational conditions and instruction. The danger has also remained recognizable: a diagnostic can be converted into ranking and control.
The 2026 release is not proof of collapse. It is a diagnosis produced by a system that remains internationally strong and has built substantial foundational knowledge. Mathematics must move from using an equation or theorem to explaining its selection and making claims that remain within the data. Japanese must move from grasping a passage to constructing an answer whose language, evidence and conditions fit together. The distance appears short. Teaching it is among the hardest tasks in education.
In a classroom that rushes toward answers, explanation becomes an attachment at the end. In a classroom built around explanation, students hear another person’s evidence, discover a gap in their own reasoning, revise and reconstruct understanding. The national test has value not on the morning its average becomes news, but when its student work changes the lesson that follows.
Sources and further reading
- NIER: 2026 National Assessment results — July 16 first release, implementation, national averages, objective and response-format breakdowns, and release schedule.
- NIER: Points from the 2026 results — Japanese and mathematics results, English IRT bands and preliminary questionnaires.
- NIER: 2026 questions, answers and explanatory material — Ratio, mimetic-word, proof, graph and data-analysis items.
- MEXT: 2026 National Assessment report and aggregate-results page — Census coverage and the three-stage publication schedule.
- ReseMom: 2026 national-test first results — Subject averages, four-skill English CBT, IRT bands and preliminary questionnaire patterns.
- Jiji Press/Nippon.com: Ratio and textual-interpretation challenges — The kippari and proportional-tape items, and warning against raw annual comparisons.
- MEXT: Overview of the National Assessment — Purpose, grades, subjects, census/sample history and individual reports.
- NIER: National achievement-test archive, 1956–1966 — Postwar purpose, subjects, sample rates and the 1961–64 junior-high census.
- National Diet Library: National testing and the Supreme Court — Bibliographic record on the Asahikawa test litigation and 1976 judgment.
- MEXT: Origins of the modern national assessment — PISA/TIMSS 2003, Japanese constructed responses, junior-high mathematics and the 2007 start.
- MEXT: PISA 2003 summary — Japan’s mathematics, reading and science positions and motivation findings.
- MEXT and NIER: PISA 2022 highlights — Scores and ranks, CBT, trend design and real-life mathematics findings.
- MEXT: 2017–19 curriculum-guideline revision — Three competency pillars, language, mathematics and active, interactive, deep learning.
- MEXT: Current curriculum and ICT — Implementation dates, GIGA School, individualized and collaborative learning.
- MEXT: Future CBT implementation, June 2026 revision — Direction for fiscal 2027 onward and longitudinal analysis.
- MEXT: Cancellation of the 2020 national assessment — COVID-19 disruption and the decision not to administer the survey.
Editor’s note: This report reflects official material and reporting available by 10:37 a.m. JST on July 17, 2026. The July 16 publication was the first stage of national averages; full national-data analysis is scheduled for August 3 and prefecture/designated-city analysis for the fall. Because questions and difficulty change, annual percentage-correct figures do not by themselves show rising or falling achievement. “Writing” is a Japanese curriculum domain; “constructed response” is an answer format, and the two are not identical. The English mean of 499 is an IRT scale score, not 49.9% and not a pass mark.