PISA scores hit record lows as family engagement also declines, study finds

Convert the OECD's own yardstick first. Thirty-five science points set against more than a year and a half of learning ^[6] puts a school year at no more than roughly 23 PISA points ^[17]. That is the exchange rate worth holding on to for every other figure in the release.

Those 35 points come from a comparison between students, not within them: two groups of 15-year-olds in one cross-section, one whose relatives ask weekly what happened at school, one whose relatives ask less often ^[6]. The OECD adjusted for socioeconomic differences ^[6], which handles income and parental occupation. It does not handle the household that talks about school because school is going well. A teenager who is failing is a harder teenager to ask.

The release carries both trends without joining them. Family support for schooling fell against 2022 in almost every country, with the sharpest drops for boys ^[4], and developed-economy averages in maths, reading and science are the lowest in the programme's history ^[2]. What is not reported is whether the countries that lost the most engagement lost the most points ^[20]. Andreas Schleicher, who runs the assessment for the OECD, describes parents seeing themselves "more as a client of the school, not as someone who takes personal interest" ^[5]. That is a diagnosis, and the published results do not include the country-by-country pairing needed to test it.

The device findings behave better, because something varies that governments actually set. More than a quarter of students in high-income countries said classmates were distracted by devices in most or every science class, and distraction rates were lower where school phone policies are strong ^[7]. In Japan, Korea and mainland China, fewer than 10% complained ^[8], a gap of more than 2.5 times ^[19]. The dose matters too: around two to three hours of in-school device use went with higher scores, and scores fell above that ^[9]. Schleicher separates leisure use, where the association with lower scores is strong, from learning use, where he calls the picture more nuanced ^[10].

Almost half of students in developed economies now use AI chatbots for learning at least weekly, and higher use went with lower scores, with never-users ahead ^[14]. Selection is the obvious rival explanation here, and a cross-section cannot rule it out: the student who is behind has the most reason to open a chatbot.

The American numbers need their own asterisk. US maths matched the OECD average and the US placed in the top 15 for science, reading and the new computational problem-solving scale, helped by other countries falling ^[12], while its reading and maths scores hit their lowest since PISA began in the early 2000s ^[3]. That decline rests on a sample that missed PISA's school and student participation standards, which the OECD flagged directly ^[13]. The baseline comparison is less fragile: 65% of US students reached baseline in maths and 74% in reading, against nearly 100% in the participating Chinese provinces ^[11], so about 35% and 26% of US 15-year-olds fall below it ^[18]. Tracey Burns of the National Center for Education and the Economy puts the stakes in workforce terms: "To win a technological race, you need the people with the science and the technology and the reading capabilities" ^[16].

What the household finding cannot yet support is a spending decision. The question PISA asks is a frequency question, and no one has randomised the asking. Until someone changes how often families ask and measures what happens to scores, the 35 points describe a gap between two kinds of household. They do not describe a return on an intervention.