Spain has just received a troubling bit of news: 15-year-old students assessed by PISA in 2025 performed worse in reading and mathematics than those tested in 2022. In science, however, the score remained statistically stable.
It is tempting to translate that snapshot into the idea that Spanish teenagers know less than three years ago. But how do we determine how much a generation knows? Can a few hours of testing measure something so broad?
The question becomes even stranger when you look behind the scenes. PISA does not test every 15-year-old. It also does not deliver exactly the same questions to all participants. Yet its data allow comparisons across countries and over time. To truly understand what that contrast means, one must first unpack the most problematic word of all: “know.”
PISA does not attempt to measure all that a teenager knows
The Programme for the International Student Assessment (PISA) does not aim to uncover everything a student stores in their head. The OECD, the body responsible for the evaluation, analyzes knowledge, skills, and attitudes, placing particular emphasis on the ability to apply what has been learned to real-life situations and problems.
Analiza conocimientos, destrezas y actitudes, poniendo especial atención en la capacidad de aplicar lo aprendido ante situaciones y problemas.
A person can remember many formulas and still stumble when deciding which one applies to a real-world case.
In mathematics, reaching at least level 2 implies, among other things, recognizing how to represent a simple situation, such as comparing the total distance of two routes or converting prices into another currency. In reading, it entails identifying the main idea of a moderately long text and locating specific information.
Therefore, a score is not a universal counter of knowledge, but describes performance in a set of competencies defined by a specific instrument. That distinction will be essential when we revisit the assertion “they know less.”
Una puntuación no constituye un contador universal del saber, sino que describe el rendimiento en unas competencias definidas mediante un instrumento concreto.
How to examine nearly 460,000 young people without seating them all in front of a screen
In Spain, 29,967 students from 956 schools participated in PISA. Statistically, that group corresponds to about 459,800 fifteen-year-old enrolled adolescents.
How can a fraction speak for a crowd? Think of a blood test. To learn certain properties of roughly five liters circulating in an adult, you don’t need to draw it all. A small, properly obtained sample suffices. With a population, something conceptually similar occurs, though the statistical operation is considerably more complex.
PISA selects schools and students through sampling procedures designed to reflect the characteristics of the target population. Then, it uses weights: some participants statistically carry more weight than others depending on how many similar people they represent and how the selection was designed.
What matters, therefore, is not interviewing half a million individuals. The key is that the sample accurately reflects the whole and that the margin of uncertainty derived from observing only a part is calculated. If the selection were biased — for example, if it systematically excluded certain centers — simply increasing the number of participants would not by itself correct the problem. Quality depends as much on design as on size.
And not everyone even answers the same questions
The surprise continues. Each participant faced about two hours of testing and was evaluated in two subjects. The youths received questions and combinations of different areas: for example, science followed by reading, or mathematics followed by science.
This prevents treating PISA as a standard classroom exam where every class gets the same booklet and then the scores are ranked. To place information from different exercises on a common scale, statistical models are used that relate observed performance to task difficulty and other task characteristics.
Here appears a somewhat unintuitive expression: “plausible values.” They are not ten alternative scores that the system assigns to each student because it does not know which one to pick. They represent possible levels of competence compatible with their answers; they are used to estimate properties of groups, not to diagnose an individual with precision.
Imagine a slightly blurred photograph. Pointing to a single pixel and claiming we know the exact position of an object would be misleading. Instead, we can delimit where it is reasonable to place it and keep that margin in subsequent calculations. Plausible values perform a comparable function.
Their usefulness becomes apparent when studying groups: when combined appropriately, they help reconstruct the distribution of capabilities without pretending to have individual-level accuracy that the test cannot offer.
Representan posibles niveles de competencia compatibles con sus respuestas, y se usan para estimar propiedades de grupos, no para diagnosticar con precisión a un individuo.
So, what does it mean that Spain has fallen?
We can return to 2025 now. The OECD concludes that Spain’s average performance was lower than in 2022 in mathematics and reading, while in science it hardly varied. Compared with 2015, all three areas show lower results and are among the country’s historical lows.
There is another important detail. The published figures are estimates, not exact measurements. When studying a sample, there is sampling error. In addition, a limited set of exercises introduces measurement error. That is why confidence intervals appear—ranges that express the uncertainty around the calculated value.
Here enters “statistical significance.” In simple terms, a meaningful difference is one that is difficult to explain only by the expected fluctuations if the populations compared did not actually differ.
Las cifras publicadas son estimaciones, no mediciones perfectas: al estudiar una muestra, existe el error de muestreo, y un conjunto limitado de ejercicios introduce el de medición.
Also, beware of translations that are too literal of PISA points into “months lost” or “years of delay.” The OECD provides approximate equivalences to contextualize certain gaps—in its 2025 briefing, for example, relating 9 points to about half a school year—but we are not dealing with a physical law.
What PISA can say and what we add
With these pieces on the table, we can separate three statements that are often mixed together.
The first is solid: a sample designed to reflect Spain’s 15-year-olds in 2025 achieved a certain performance in the evaluated competencies, and that data allows a statistical comparison with earlier generations.
The second needs more care. A decline compatible with the program’s statistical criteria provides evidence of a deterioration in the concrete abilities the assessment aims to capture. It does not amount to inventorying everything learned by each adolescent.
The third crosses a boundary that data alone cannot cross. PISA does not demonstrate that every Spanish teenager is less intelligent, that they have forgotten more content, that every dimension of their education has worsened, or that a particular education law is responsible for the evolution.
Those who participated in 2025 are not the same adolescents who were tested in 2022. Each edition takes a new sample of similarly aged people. We compare equivalent cohorts, not observe how the background of the same individuals wanes over time.
No equivale a inventariar todo lo aprendido por cada adolescente ni demuestra que sea menos inteligente, que haya olvidado más contenidos, que toda dimensión de su educación haya empeorado o que una ley educativa determinada sea la responsable.
That is why “the Spanish students know less” functions as a media summary, but it enlarges what the numbers actually attest. The precise formulation would be less sensational: a new generation shows poorer performance in certain competencies.
A thermometer can detect fever, but not diagnose the illness
Another common leap remains: moving from confirming that something has changed to deciding why it happened. Here a thermometer helps. If it reads 39 degrees, it provides valuable information and can alert us to a problem; but the thermometer, by itself, does not identify the fever’s cause.
With PISA, something similar happens. Screens, social media, the pandemic, inequality, reading habits, artificial intelligence, school resources, teaching staff, or legislative reforms can become hypotheses. Finding that one characteristic coincides with higher or lower scores allows us to investigate relationships, but an association does not automatically prove causation.
The report itself illustrates the risk. In Spain, those who reported digital distractions during science classes scored 16 points lower than those who did not report them, after accounting for students’ socioeconomic profile and the schools. That is not enough to claim that devices caused exactly that loss.
And the easy tale of blaming only COVID does not fit with the international series either. The OECD emphasizes that the decline in many countries began before 2020.
So, do Spanish students know less?
The answer depends on what we mean by “know.”
PISA cannot certify that Spanish teenagers possess less knowledge in any imaginable sense. It does, however, provide statistical evidence that the generation evaluated in 2025 performed worse than the 2022 generation in reading and mathematics within the competencies that the program measures. In science, that pronounced decline did not occur.
Just as a blood sample can reveal certain properties of an organism without emptying its veins, a representative selection reveals characteristics of a vast population without testing every individual.
But it is worth remembering the thermometer too. A good measurement can alert us that something is moving and estimate the magnitude of the change without automatically diagnosing its origin.
PISA no puede certificar que los adolescentes españoles posean menos conocimiento en cualquier sentido imaginable; sí proporciona evidencia estadística de que la generación evaluada rindió peor en las competencias que mide.
The takeaway from PISA goes beyond the classroom: numbers can describe a part of reality with extraordinary sophistication, but understanding what they mean also requires recognizing where the things they measure end.
