Are Students Getting Worse at Maths and Reading? What Three Assessment Systems Say — and What Explains the Fall

The question

Headlines cycle between two poles: a panic that smartphones are making children measurably dumber, and a reassurance that tests are simply harder, or kids less motivated to try, or the comparisons flawed. The underlying empirical question is narrower and more answerable than either pole suggests: do the large international assessment systems actually show student skills declining, and if so, how much of the decline can be attributed to identified causes?

Three assessment systems answer the measurement question independently, with different samples, different age groups and different test designs: PISA tests 15-year-olds in reading, mathematics and science every three years; TIMSS tests fourth- and eighth-graders in mathematics and science; PIRLS tests fourth-grade reading. When all three move the same direction, sampling noise or a single instrument's quirks are not plausible explanations.

This brief takes the question in three steps: what is actually falling, what the pandemic explains, and what the fashionable suspects — phones, screens, the culture of reading — do and do not account for.

Why it matters

Educational achievement at the national level is a slow-moving input to productivity, inequality and civic capacity. A decline of 20–30 points on the PISA scale — roughly a year of schooling, by OECD's own conversion — sustained across a decade, implies meaningful long-run costs if it reflects real skills rather than test artifacts. It also compounds: skills lost in grade 4 shape grade 8 outcomes, which shape post-secondary sorting. And the distributional pattern matters as much as the mean: in the latest data the bottom of the distribution is falling faster than the top, which widens inequality precisely when education systems claim to be narrowing it.

What is actually falling — and what is not

The newest evidence point is PISA 2025, released 8 September 2026 — and it removes any doubt that the 2022 drop was a one-cycle anomaly. Across the 35 OECD countries with comparable data, mean mathematics fell about 9 score points and reading about 14 points between 2022 and 2025, while science did not change significantly on average. Over the decade 2015–2025, OECD averages fell from 485 to 463 in mathematics (−22 points, "just over one year of learning" in OECD's framing, using the benchmark of 20 points per school year) and from 489 to 461 in reading (−28 points); science fell modestly, 489 to 482. Among non-OECD participants science was broadly stable (436 to 437) while mathematics (434 to 426) and reading (428 to 412) also declined. OECD's summary is blunt: following the unprecedented drop recorded in PISA 2022, "learning outcomes have largely stagnated or continued to decline in many countries."

The shape of the decline matters more than the headline. Four regularities recur across the systems:

• Reading fell first, mathematics later. OECD average reading peaked around 2012 and has been falling since; mathematics was broadly flat from 2003 to 2018 and then dropped sharply (−15 points in the 2018–2022 cycle, described by OECD at the time as three times greater than any previous consecutive-cycle change, then −9 more by 2025). The reading decline over 2018–2025 totals −25 points — students in 2025 performed at the reading level that 14-year-olds would have shown seven years earlier.
• The bottom falls faster than the top. In PISA 2022–2025 mathematics, the decline was steeper at the 10th percentile than at the 75th (−7) or 90th (−5), and the 10th-to-90th percentile gap widened by about 8 points in mathematics and 3 in reading. The United States in TIMSS 2023 is the extreme case: fourth-grade mathematics at the 10th percentile fell 37 points from 2019 — the lowest in the assessment's history back to 1995 — while the 90th percentile did not move measurably, and fourth- and eighth-grade mathematics averages fell 18 and 27 points respectively. The NCES Commissioner's summary: "Globally, we are seeing declines in achievement, even in traditionally high-performing systems."
• Science is more resilient on average but not unscathed. PISA 2025 science means were flat 2022–2025, yet the science performance gap between highest and lowest achievers widened by about 3 points in that window and by more than 15 points since 2018, and the share of low performers in science rose (from 16% in 2022). Flat means can mask distributional deterioration.
• The declines are widespread but not universal. In PIRLS 2021, reading achievement fell from 2016 to 2021 in 21 of the 32 countries with comparable data (8 unchanged, 3 higher). In TIMSS 2023 mathematics, grade 8 fell in 14 of 34 comparable countries against 3 that rose, while grade 4 was more balanced (14 up, 13 down, 22 unchanged of 49). Yet some systems are rising: science scores increased in Costa Rica, the Slovak Republic, Türkiye and the United Kingdom among OECD members, and in a dozen non-OECD systems from Cambodia to Uruguay, and top performers (Singapore, Chinese Taipei, Estonia, Japan, Korea) remain stable or improving. And Brazil stands out in the other direction: OECD reports Brazil's performance stable across all three domains since 2015 while assessment coverage rose four percentage points.

Could it be measurement? The comparability case is strong but not unlimited — this series' brief on cognitive-test comparability covers the general problem; here are the numbers that apply directly. PISA's own link errors — the uncertainty in putting two cycles on one scale — are 2.24 points for 2018–2022 mathematics, 1.47 for reading and 2.9 for 2022–2025 science, far smaller than the declines. Changing student composition explains a small slice: rising immigrant shares (correlated with a 40-point achievement gap) account for roughly 1.5 of the 25 OECD reading points lost since 2018. TIMSS 2023 completed its transition to digital delivery (begun in 2019), a mode change that demands caution for individual 2019–2023 comparisons but cannot manufacture a coordinated multi-country pattern. The clean summary: the declines are too large, too cross-instrument and too distributionally structured to be artifacts — though every individual country estimate carries more uncertainty than headlines imply.

The pandemic: a real shock, not the whole story

The 2020–2021 school closures are the most-identified single cause in this literature, and the evidence is unusually good for education economics.

• The gold-standard natural experiment: Engzell, Frey and Verhagen (2021) exploited the Dutch lockdown — only eight weeks, in a wealthy, equitable, fully broadband-connected system — comparing ~350,000 pupils' national test progress against the same period in the three prior years. Learning loss: about 3 percentile points, or 0.08 standard deviations — one-fifth of a school year, exactly matching the closure duration — and up to 60% larger among children from less-educated homes. The effect reflected cumulative knowledge loss, not test-day disruption.
• The cross-country dose-response: Jakubowski, Gajderowicz and Patrinos (2025), working with PISA data covering 72 countries and roughly 175 million 15-year-olds, estimated that the pandemic mathematics decline averaged about 0.14 SD, with short closures associated with 8–10 point losses and longer ones with 15 or more, and larger losses for boys, immigrant and disadvantaged students.

These findings are real and regressive. But they cannot carry the whole explanation, for three reasons.

First, timing. Reading had been falling for a decade before COVID-19; mathematics was flat from 2003 to 2018. The pandemic arrived in 2020.

Second, OECD's own reading of the 2022 and 2025 cycles. In the PISA 2022 companion volume, the decline was described as only partially attributable to COVID-19, with no clear trend difference between systems that closed schools briefly (Iceland, Sweden, Chinese Taipei) and those that closed them for long periods (Brazil, Ireland, Jamaica). The 2025 volume goes further: "The weakening of performance predates the COVID-19 pandemic in many countries, pointing to deeper and more persistent challenges within education systems."

Third, the post-reopening window. If closures were the whole story, the 2022–2025 period — schools open everywhere — should have shown stabilization or rebound. Instead mathematics fell another 9 points and reading 14 across the OECD. A one-time level shift followed by a resumed slope is exactly what the data show, and only the level shift is the pandemic's to claim.

The defensible decomposition: closures caused a measurable, unequal, cumulative loss — plausibly the bulk of the 2016/2018–2021/2022 drop — layered on top of a decline that was already underway and has continued since.

Phones and digital life: the favourite suspect fails key tests

The smartphone explanation is popular, intuitively satisfying, and performs poorly against the evidence on both of its required legs: the timeline and the intervention.

The timeline leg fails for mathematics. Smartphone saturation of adolescent life happened across the 2010s. PISA mathematics was flat from 2003 to 2018 across the OECD — the entire adoption decade — and only began its steepest fall after 2018. A slow-acting environmental toxin does not leave a flat trend for fifteen years and then triple its slope. For reading the timeline is more favourable (the decline starts around the 2012 peak), which is why the screen-time hypothesis retains plausibility for reading specifically — but a hypothesis that explains one domain and misses the other is not a general account of falling scores.

The intervention leg fails on results. If phones were a primary cause of achievement decline, banning them in schools should show measurable gains. The two best studies find essentially nothing on the average achievement margin. Abrahamsson's event study of Norwegian smartphone bans (published in the Journal of Human Resources in 2026) found no average effect on education or mental health, but important composition effects: girls' GPA and track placement improved — most among the disadvantaged — psychological consultations fell, and bullying dropped for both genders, while boys' outcomes did not move. The SMART Schools study in England (Lancet Regional Health – Europe, 2025) compared 1,277 pupils across 30 schools with restrictive versus permissive phone policies: restrictive policies reduced phone and social-media use during breaks, but wellbeing showed no difference (adjusted mean difference −0.48, 95% CI −2.05 to 1.06) and attainment showed no association. A policy that demonstrably changes the behaviour it targets, yet does not move test scores, is evidence against that behaviour being a main driver of test scores.

What the correlational evidence does support. PISA itself reports that more than one in four students say classmates are distracted by digital devices in most or every science lesson (the 2022 cycle found 30% in most mathematics lessons), that excessive device use and digital distraction are associated with weaker outcomes, and — the nuance usually dropped — that moderate device use for learning is associated with stronger science performance. OECD's 2025 reading of its own data is careful about causality: "In reading, increased digitalisation, longer screen time and falling rates of reading for enjoyment have coincided with weaker literacy outcomes." Coincided, not caused. The honest status of phones: a real source of classroom distraction and a plausible contributor through attention and reading-time displacement; unsupported as the main cause, and the policy lever the public reaches for first (banning phones at the school door) is not the lever the evidence recommends expecting much from.

Reading culture and motivation: the leading secular channel

If the pandemic explains a chunk of the recent drop and phones fail as the general explanation, what tracks the decade-long reading slide? The best available answer is the measured collapse in reading as a leisure activity.

Bone and colleagues (iScience, 2025) document a 20-year decline in reading for pleasure in the American Time Use Survey. OECD's 21st-Century Readers (2021) had already documented across countries that both enjoyment of reading and time spent reading for leisure were falling — through 2018, i.e., before the pandemic. PIRLS data consistently show reading engagement and enjoyment associated with reading achievement. And the OECD's PISA 2025 volume puts digitalisation, screen time and falling reading enjoyment in the same explanatory paragraph as the reading decline.

The mechanism is plausible: reading skill is use-dependent, leisure reading is where adolescents get most of their volume, and screens captured the leisure-hours margin that books used to occupy. But the evidence grade must be stated plainly: this is a correlation across countries and time between falling reading habits and falling reading scores, with a credible mechanism — not an identified causal estimate. No natural experiment has yet shifted population-level reading-for-pleasure and measured the achievement consequence. The claim "less reading causes weaker reading" is about as defensible as causal claims get without an experiment, and still weaker than the pandemic estimates.

What else is on the table

Several candidates remain live, none yet decisive:

• Adolescent mental health, sleep and school engagement. TIMSS 2023 reports roughly one in ten students internationally absent weekly, and disengagement correlates with achievement; PISA 2025 tracks declines in students' perceived engagement alongside performance. These are correlates and candidate mechanisms, not identified causes.
• School systems themselves. Curriculum churn, teacher shortages and post-pandemic disruption to instruction quality are plausible, heterogeneous, and largely unstudied at the cross-country identification level.
• Environmental exposures. Small causal effects are documented for specific exposures — this series' air-pollution brief, for instance, finds well-identified test-day cognitive penalties from pollution — but nothing in that literature approaches the magnitude needed to move national means by 20–30 points.
• Pure measurement artifact. Ruled out as the main story by the link-error arithmetic, the cross-instrument agreement and the distributional structure of the declines. A useful reality check comes from an unexpected direction: Bratsberg and Rogeberg (PNAS, 2018) showed that measured IQ rose and then fell across Norwegian conscript cohorts born 1962–1991 — within the same families. Whatever else that finding means, it establishes that real environmental changes can move measured cognitive ability downward within one generation, and that compositional or genetic explanations are not required. Measured ability can genuinely fall.

The best current answer

Are measured student skills falling? Yes — with high confidence. Three independent assessment systems, link errors far smaller than the declines, and a distributional structure that no artifact hypothesis reproduces.

What explains it? A layered answer:

1. Pandemic school closures: a real, regressive, quantified shock — roughly 0.08 SD after eight weeks in the best-isolated case, with cross-country dose-response — accounting for much of the 2016/2018–2021/2022 drop. Confident.
2. A pre-existing secular decline, strongest in reading, running since roughly 2012, whose leading measured correlate is the collapse in reading for pleasure and reading enjoyment, plausibly mediated by screen-displaced leisure time. Suggestive, not identified.
3. Phones as the main cause: not supported. Flat mathematics through the adoption decade; null average effects in the best ban quasi-experiments; OECD's own language stays correlational. Phones are a contributor to distraction, not the engine of the decline.
4. A residual nobody has explained — including the continued 2022–2025 declines after reopening. The intellectually honest version of the popular headline is: students are measurably getting worse at maths and reading, the pandemic did a chunk of it, phones probably did not, and we do not know the rest.

Evidence quality, caveats, and what would change the picture

• The measurement evidence is much stronger than the causal evidence. The decline is about as solid as social-science facts get; the decomposition is not. Any presentation that reverses that asymmetry — confident cause stories, hedged measurement — misrepresents the state of knowledge.
• PISA measures 15-year-olds' applied skills, not IQ and not pure "ability." Conclusions transfer to schooling-relevant competence more readily than to underlying cognitive capacity; the Norwegian conscript result shows the two can move together, not that they always do.
• Country-level attribution is weak by construction. Cross-country regressions on closure length, device penetration or reading habits are ecological and confounded; the credible causal estimates (Dutch lockdown, Norwegian bans, English phone-policy comparison) are national natural experiments answering narrower questions.
• Publication and political incentives run toward alarm. Declines make headlines; flat trends and rising systems (Singapore, Estonia, the Slovak science reversal) get less attention, and this brief has deliberately included them.
• What would update this most: the next PISA cycle's trajectory (does the post-2022 slope continue, flatten or revert?); longer follow-up of phone-ban quasi-experiments into adulthood; and any design that shifts population-level reading-for-pleasure independently of achievement trends.

Verdict and calibrated confidence

That measured mathematics and reading skills are genuinely falling across most rich countries: ~92%. Three instruments, small link errors, consistent distributional structure, no competing artifact account.

That pandemic closures caused a real learning loss and explain much of the 2016/2018–2021/2022 drop: ~88%. A clean natural experiment plus a cross-country dose-response, with limitations (OECD's own closure-comparison puzzle) that argue for "much of," not "all of."

That the pandemic alone explains the declines: ~10%. Pre-2018 trends and the continued 2022–2025 slide make a single-shock account untenable.

That smartphones and social media are the main cause: ~15%. Fails the mathematics timeline and the ban experiments; retains plausibility as a partial contributor to the reading channel.

That the decline in reading for pleasure is a genuine causal contributor to the reading slide: ~60%. Right sign, right timing, right mechanism, cross-country correlation — no identification.

That known causes fully account for the decline: ~25%. The honest residual is large. Anyone claiming a complete story is outrunning the evidence.

What would move me most: a PISA 2028 showing stabilization would push the residual toward pandemic-plus-recovery; another decade of continued decline through 2030 would push it toward slow cultural or institutional causes still poorly measured. On present evidence, the defensible summary is: the decline is real, the pandemic did a measurable share of it, the famous culprits underperform, and the deepest cause of the rest is genuinely unknown — which is a more useful place to stand than either manufactured panic or manufactured reassurance.

References

1. OECD. PISA 2025 Results (Volume I): Future-Ready Students. PISA. OECD Publishing, Paris; 2026. doi:10.1787/73451bc5-en. https://www.oecd.org/en/publications/2026/09/pisa-2025-results-volume-i_5265bfb1.html
2. OECD. PISA 2022 Results (Volume I): The State of Learning and Equity in Education. PISA. OECD Publishing, Paris; 2023. doi:10.1787/53f23881-en. https://www.oecd.org/en/publications/pisa-2022-results-volume-i_53f23881-en.html
3. OECD. PISA 2022 Results (Volume II): Learning During — and From — Disruption. PISA. OECD Publishing, Paris; 2023. doi:10.1787/a97db61c-en. https://www.oecd.org/en/publications/pisa-2022-results-volume-ii_a97db61c-en.html
4. OECD. PISA 2022 Results (Volume I and II): Country Notes — Brazil. OECD Publishing, Paris; 2023. https://www.oecd.org/en/publications/pisa-2022-results-volume-i-and-ii-country-notes_ed6fbcc5-en/brazil_61690648-en.html
5. Mullis IVS, von Davier M, Foy P, Fishbein B, Reynolds KA, Wry E. PIRLS 2021 International Results in Reading. Boston College, TIMSS & PIRLS International Study Center; 2023. doi:10.6017/lse.tpisc.tr2103.kb5342. https://pirls2021.org/results/trends/overall/
6. von Davier M, Kennedy A, Reynolds K, Fishbein B, Khorramdel L, Aldrich C, et al. TIMSS 2023 International Results in Mathematics and Science. Boston College, TIMSS & PIRLS International Study Center; 2024. doi:10.6017/lse.tpisc.timss.rs6460. https://timss2023.org/results/grade-8-math-achievement-trends/ and https://timss2023.org/results/grade-4-math-achievement-trends/
7. National Center for Education Statistics. Mathematics scores of U.S. fourth- and eighth-graders decline on international mathematics and science assessment. U.S. Department of Education; 4 December 2024. https://nces.ed.gov/learn/press-release/mathematics-scores-u-s-fourth-and-eighth-graders-decline-international-mathematics-and-science
8. Engzell P, Frey A, Verhagen MD. Learning loss due to school closures during the COVID-19 pandemic. Proceedings of the National Academy of Sciences. 2021;118(17):e2022376118. doi:10.1073/pnas.2022376118.
9. Jakubowski M, Gajderowicz T, Patrinos P. The COVID-19 pandemic and student achievement losses: cross-country evidence. npj Science of Learning. 2025. doi:10.1038/s41539-025-00297-3.
10. Bratsberg B, Rogeberg O. Flynn effect and its reversal are both environmental in origin. Proceedings of the National Academy of Sciences. 2018;115(26):6774–6779. doi:10.1073/pnas.1718793115.
11. Abrahamsson S. Smartphone bans, student outcomes and mental health. Journal of Human Resources. Published 11 March 2026. doi:10.3368/jhr.0224-13403r2.
12. Goodyear VA, Kessel AS, Fletcher T, et al. School phone policies and adolescent wellbeing and attainment: the SMART Schools cross-sectional study. The Lancet Regional Health – Europe. 2025. doi:10.1016/j.lanepe.2025.101211.
13. Bone Z, et al. The decline in reading for pleasure over 20 years of the American Time Use Survey. iScience. 2025;28:113288. doi:10.1016/j.isci.2025.113288.
14. OECD. 21st-Century Readers: Developing Literacy Skills in a Digital World. OECD Publishing, Paris; 2021. doi:10.1787/a83d84cb-en.