How to Read a Bad Data Point: PISA 2025

On Tuesday, September 8, first thing in the morning, someone left their mobile phone on the staff room table with the headline facing up. Marta, principal of a three-form entry school, read it standing up, her coffee untouched, while around her that particular silence of bad grades settled. This notebook had described the scene a few weeks earlier, almost with anticipated nostalgia: the PISA liturgy, front pages featuring the word "collapse, " talk shows looking for scapegoats before lunch. Only the calendar failed. The liturgy did not wait for December; it arrived in the first week of term, with books still uncovered, because the OECD was launching a new cycle and Spain debuted a new disappointment.

Let us start with the numbers, which almost no one has presented without adjectives. According to the OECD's own official indicators, Spain obtained an average score in PISA 2025 of 477 in science, compared to an OECD average of 482; 457 in mathematics, compared to 463; and 451 in reading, compared to 461. Compared to 2022, science held steady (the organization itself does not consider the change statistically significant), mathematics fell by sixteen points, and reading fell by twenty-three. There lies the data that deserves a serious conversation: the reading comprehension of fifteen-year-old Spanish students has been eroding for years, and the pace is not slowing down. The OECD itself places these results among the lowest recorded to date in Spain in all three subjects, and the average across OECD countries has reached its lowest level since PISA began; for the purpose of this piece, what has been verified is enough, and indeed it is quite a lot.

What the thermometer says

Anyone who has read "What PISA measures and what it doesn't" here knows the house formula: PISA is a reasonable thermometer and a terrible judge. Well, the thermometer has registered a fever, and honesty obliges us to state it in full. In national aggregate and long-term trend (the only use the instrument can actually withstand), a twenty-three point drop in reading over three years is neither statistical noise nor a survey conspiracy. It reports something real that any secondary school teacher recognizes through other means: students for whom a page-and-a-half text is too long, vocabulary that is narrowing, growing difficulty in sustaining a written argument. Dismissing the data with "PISA doesn't measure everything" would be the symmetrical trap that this piece intends to dismantle. True, it does not measure everything. What it does measure, it measures reasonably well, and this time it has measured an erosion.

It is worth adding, for hygiene, that the fever is not a Spanish anomaly. The OECD average is also declining, and the underlying trend in science has been downward for all countries for a decade. This offers no consolation, but it does provide direction: what the exam portrays does not seem to be the failure of a specific law in a specific country, but rather a change of era in the conditions of learning that affects very different educational systems alike.

The second wave

The first wave of a PISA result is the headlines. The second, more interesting and more damaging, is the allocation of blame. It has been underway for ten days: everyone has read in the table the confirmation of what they already thought on the 7th, and the education press publishes surveys literally asking if PISA proves that we are going wrong, with the "we" conveniently open-ended so that each reader can insert their favorite adversary. Autonomous communities are compared with decimal-point rigor and tons of intent. Drops are attributed to other people's pedagogy, whatever one's own may be.

This notebook is not going to play that game, and not out of decorative equidistance. The fine print of the report itself warns that differences between systems are largely explained by socioeconomic factors that the exam only partially adjusts for; the statisticians who audited the psychometric model have shown that rankings shift depending on the assumptions adopted. An instrument designed to take the temperature of entire systems cannot pass judgment on a region, a school type, or a methodology, for the simple reason that it does not isolate their effects from everything else that happens in a student's life. Using it as a judge is cheating with the appearance of rigor. That others cheat does not oblige any faculty to imitate them.

An exam in the middle of a move

There is also a circumstance that headlines barely capture: this exam has portrayed a classroom in transition. The PISA 2025 context questionnaire, according to the OECD country note, describes a Spanish school system that has massively restricted mobile phones, with 86% of students in schools that do not allow them on school premises (67% in 2022; 49% on average across the OECD), and an adolescence that turns to artificial intelligence chatbots to learn at least once a week in more than half of cases (54%, compared with 46% on average across the OECD). Between these two figures lies almost all the pedagogical bewilderment of the moment.

The temptation is to recruit this data into one of the two customary churches, that of panic or that of the digital gospel. It is better to stick with what the organization itself sets out in the report summary, something that should be framed in more than one department meeting: students who use AI for specific schoolwork tasks, such as summarizing, drafting or researching, score around 20 points lower in science on average than students who do not use it, while weekly users of AI for general purposes perform similarly to non-users, and the relationship, the OECD stresses, is complex and depends on how it is used. It is an association, not a verdict, but it fits the doctrine that this notebook has been advocating since the piece on homework: a tool can improve the output of a task without any learning having taken place. The underlying problem is not the tool but the asymmetry: entire industries optimized to capture attention and substitute effort, up against an institution whose job consists precisely of shaping attention and giving meaning to effort. That struggle is not won with circulars, nor by losing faith in the classroom.

What a staff body can do this week

First, read the source and not the echo. The country note published by the OECD on Spain is only a few pages long and is within reach of any management team; half an hour of staff meeting with the original document in hand immunizes against weeks of headlines. It is an almost therapeutic exercise: checking what it really says, what it does not say, and how many of the claims heard these days do not actually come from there.

Second, keep the signal and let go of the noise. The signal is reading. If the aggregate data is of any use to a specific school, it is to ask calmly what place sustained reading occupies in its daily life (not on paper, but in the schedule), how much long text a student actually gets through per week, and what the school does with attention, which is the gateway to everything else. This notebook has dedicated entire pieces to that garden; a bad PISA result is a good excuse to water it.

And third, do not let the ranking rewrite the school project. A school with an educational vision does not educate for the percentile, and families do not entrust their children to it because of Spain's position on a table. It educates for things that the exam does not visit (discernment, character, a taste for what is worthwhile) and that requires staff bodies with their own depth, capable of reading a table without trembling and without shrugging their shoulders. In that arena is where Nomba Education works: giving teachers the cultural depth from which data is read with discernment instead of endured with fright. Marta, with her untouched coffee, does not need an emergency plan for December. She needs a staff body that knows what the thermometer measures, what it does not measure, and what they were already doing, with or without a fever, for the minds of their students.

Cover image generated with artificial intelligence.