PISA Guide: Test, Scores, Rankings & How It Works

PISA stands for the Programme for International Student Assessment. It is organised by the Organisation for Economic Co-operation and Development (OECD) and compares what 15-year-old students know and what they can do with that knowledge.

The assessment is age-based rather than grade-based. For PISA 2025, the target population consisted of students generally between 15 years and 3 months and 16 years and 2 months at the beginning of the testing period who were enrolled in Grade 7 or higher. Students can therefore be in different grades, school types or educational tracks while still belonging to the same PISA target population.

PISA is not designed mainly to test whether students remember a particular national curriculum. Tasks ask students to interpret information, reason through unfamiliar situations, evaluate evidence and apply knowledge to problems. This allows systems with different curricula to be compared on common assessment scales.

Students are selected through sampling

PISA does not test every eligible student. In PISA 2025, countries and economies used a two-stage sampling design. Schools with eligible 15-year-olds were selected first. Students were then sampled within those schools.

For computer-based participants, the standard target was generally 42 students from each sampled school when enough eligible students were available. Statistical weights are later used so the participating students represent the wider population covered by the assessment.

Most students take a computer-based assessment

Most participating systems used computer-based testing in PISA 2025. The cognitive assessment lasted about two hours. Students also completed a background questionnaire of about 35 minutes covering their home circumstances, attitudes, learning experiences and school environment. School principals completed a separate school questionnaire.

Core-domain questions included both selected-response items and questions requiring students to construct their own responses. Questions were commonly organised around a scenario, problem, text or real-life situation.

The test is adaptive

PISA 2025 used a multi-stage adaptive approach for computer-based science, mathematics and reading. The blocks of items presented later in the assessment could depend on performance in earlier blocks.

Different students therefore did not need to answer exactly the same questions. Statistical scaling connects the different test forms to common PISA scales, allowing population-level results to be estimated from the combined assessment design.

One of the three core domains receives greater assessment coverage in each cycle. Science was the major domain in 2025. Mathematics had been the major domain in 2022, while reading was the major domain in 2018.

PISA 2025 also included Learning in the Digital World and an optional English-as-a-foreign-language assessment. Volume I of the 2025 results reports the three core domains and computational problem solving. Further PISA 2025 volumes covering additional assessments are scheduled separately.

PISA points are scale scores, not percentages. A score of 500 does not mean that a student answered 50% of the questions correctly, and a country mean of 500 does not mean that every student in that country performs at the same level.

The scales were constructed so results were distributed around approximately 500 points with a standard deviation around 100 points when the original scales were established. The scales remain linked across cycles, so the OECD average is not reset to 500 whenever a new assessment is conducted. This is why the PISA 2025 OECD averages were 482 in science, 463 in mathematics and 461 in reading.

Why PISA uses plausible values

Each student sees only part of the total pool of assessment material. PISA therefore uses statistical models to estimate population performance. International databases provide multiple plausible values for student proficiency rather than treating one calculated number as a perfectly measured individual score.

For population analysis, estimates are calculated across the plausible values and combined with the survey weights and variance procedures specified for PISA. Plausible values are designed for population statistics; they should not be interpreted as a collection of ordinary individual test marks.

PISA rankings are often presented online as a simple ordered list, but the statistical interpretation is more careful. Country and economy means are estimated from samples, and every estimate has uncertainty. Two systems separated by a few score points can therefore have performance that cannot be reliably distinguished with the available data.

An ordered table based on reported mean score is useful for navigation and comparison. It should be described as score-based ordering, not automatically as a set of exact, error-free ordinal ranks.

OECD rank ranges account for uncertainty

The OECD calculates confidence intervals for ranks using simulation methods. These rank ranges incorporate uncertainty in country/economy mean estimates and the multiple comparisons involved when many systems are ordered at the same time.

The PISA 2025 technical notes give a useful science example. Australia and New Zealand had the same rounded mean science score of 509, yet their estimated rank intervals were not represented by one definitive ordinal number. Australia’s 95% rank range was 8th to 14th among participating systems, while New Zealand’s was 8th to 15th.

Mean scores are useful for comparisons, but proficiency levels provide another view of performance. The PISA scales are divided into levels associated with increasingly demanding types of tasks.

For the main PISA 2025 science scale, levels run from Level 1b through Level 6. Level 2 is used as the baseline level of proficiency. Students reaching Level 2 or higher can perform a defined set of science tasks that go beyond the most elementary items. Students at Levels 5 and 6 are classified as top performers in OECD reporting.

In PISA 2025, about 74% of students across OECD countries reached at least Level 2 in science, while about 7% reached Levels 5 or 6.

The OECD average is frequently used as a reference point in PISA reporting. It is not a global average of all participating countries and economies, and it is not a fixed score of 500.

The set of OECD countries included can also depend on the analysis. A cross-sectional average for a current cycle and a trend average restricted to countries with comparable historical data can represent different comparison sets.

A system can be numerically above or below an OECD average without the difference necessarily being statistically distinguishable. OECD result tables indicate where comparisons require that distinction.

Trend analysis is one of PISA’s main uses, but comparisons have to follow the linked scales and OECD comparability rules. The strongest long-term subject series generally begin with the cycle in which that subject first became the major domain: reading in 2000, mathematics in 2003 and science in 2006.

Countries and economies have not all participated in every cycle. Reporting geography can change, assessment modes can change, data can carry sampling cautions, and some historical observations may not be comparable with later results. A trend should therefore use the actual official observations available for the same system rather than filling missing cycles or assuming uninterrupted coverage.

The latest OECD results also show why cycle and trend are different concepts. Across OECD countries, science averaged 482 points in 2025, compared with 489 in 2015. Mathematics fell from 485 to 463 over the same period, while reading fell from 489 to 461. These long-period movements cannot be inferred from a single ranking table.

PISA is a cross-sectional international assessment. It does not follow the same students through their entire education, and country-level comparisons reflect many differences in student populations, institutions and social conditions. Descriptive associations from PISA data therefore require more evidence before being interpreted as causal effects.