Screens Track Lower Grades and Language; the Effect Is Small

The tablet is finally face down on the kitchen counter and the four-year-old has been asleep for an hour. Somewhere in the day a push notification arrived — 2.8 million children, screens, worse outcomes — and it has been sitting at the back of the evening ever since. The arithmetic being done at that counter is not really about the study. It is about the ninety minutes of cartoons that bought the time to cook dinner, and whether those ninety minutes are the thing that shows up later in a report card.

The headline number travelled unusually well this week. The number underneath it did not travel at all. Nearly every outlet reported the population — 2,817,252 children across 64 countries — and almost none reported how large the association actually was. That omission matters more than it sounds, because in this particular paper the size of the effect is the finding. The review is large, pre-registered and careful, and its central result is that the links between screen use and children's development are real, consistent, and very small.

The paper, published in JAMA Pediatrics on 28 September 2026 by Elizabeth Al-Jbouri, Sheri Madigan and six co-authors, is an umbrella review: not a study of children, but a study of meta-analyses. The authors searched Embase, MEDLINE and PsycINFO from each database's inception to 25 November 2025, and kept meta-analyses of population-based samples of children aged 18 years or younger published in English. Twenty-three met the bar. Together those 23 syntheses carried data on 2,817,252 children from 64 countries.

Two reviewers independently extracted data at both the meta-analytic and the primary-study level, and effect sizes from across the literature were converted into a single common currency — the Pearson correlation, r — so that findings measured on different scales could be reanalysed together. The review adhered to the PRISMA reporting guideline. The four outcome families the authors set out to test were academic functioning, cognitive skills, socioemotional development and mental health.

An umbrella review is a powerful instrument and a blunt one. Pooling 23 meta-analyses buys enormous statistical precision and a view of a whole literature rather than one corner of it. What it cannot do is improve on the studies underneath, and those are overwhelmingly observational — measuring screen use and development as they occur rather than assigning children to conditions. A review of correlations, however large, returns correlations.

Screen-based media exposure was associated with lower academic achievement, at r = −0.15 (95% CI, −0.23 to −0.07), and with weaker language skills, at r = −0.14 (95% CI, −0.18 to −0.10). On the mental-health side it was associated with more internalising behaviour — the withdrawn, anxious, low-mood cluster — at r = 0.10 (95% CI, 0.07 to 0.13) for screen use in general and r = 0.09 (95% CI, 0.06 to 0.12) for social media specifically, and with more externalising behaviour, the acting-out cluster, at r = 0.11 (95% CI, 0.08 to 0.14).

Those coefficients are where the story sits, and they need translating. A correlation of 0.15, squared, is about 0.02: screen use corresponds to roughly 2% of the variation between children in academic achievement. The language figure works out to about the same. Put the other way round, some 98% of why one child reads better than another lies somewhere other than screen time. The confidence intervals are tight and clear of zero, so these associations are almost certainly not statistical noise. They are simply small.

Two results in the review look considerably larger, and they are the two to read most sceptically. Emotional problems came in at r = 0.30 and peer problems at r = 0.28 — but with confidence intervals of 0.08 to 0.50 and 0.04 to 0.49 respectively. An interval running from 0.04 to 0.49 is an interval consistent with almost nothing and with a moderate effect alike. These are the least precise estimates in the paper, and by some distance the most quotable.

The review's null results drew almost no coverage, and there are five of them. Screen-based media exposure was not associated with executive functioning (r = 0.01; 95% CI, −0.09 to 0.11), with mathematics scores (r = −0.10; 95% CI, −0.23 to 0.03), with science scores (r = −0.03; 95% CI, −0.15 to 0.11), with reading scores (r = −0.02; 95% CI, −0.12 to 0.09), or with prosociality (r = −0.22; 95% CI, −0.43 to 0.01). In each case the interval crosses zero.

That list sits in an interesting tension with the headline result. Academic achievement was lower, yet mathematics, science and reading — three of the ways academic achievement is actually measured — did not move individually. Executive functioning, the cognitive machinery a great deal of screen-time theorising rests on, showed an association of essentially zero. A review candid enough to publish those five results alongside the six positive ones has earned being read in both directions.

The authors also report that screen type, the subtype of behaviour problem, publication year and the children's age each moderated some of the associations, which means the pooled figures above average over real differences. The published abstract does not break those moderated estimates out, so how much the picture shifts between a phone and a television, or between a toddler and a fifteen-year-old, is not something this reading can quantify.

The most consequential limit is one the design makes unavoidable: this is a body of observational evidence, and it cannot settle which way the arrow points. A correlation between screen use and lower language scores is equally consistent with screens displacing the conversation that builds language, and with children who are already behind in language being handed a tablet more often. Both stories predict the same r. Only designs this review did not have — randomisation, or careful long-run tracking within the same child — can separate them.

A 2% effect is nonetheless not a nothing, for a reason that has little to do with any individual household. Spread across a population, a small shift in an average moves the tails, and at the scale of 64 countries a correlation of 0.15 stands for a meaningful number of children. That is precisely why the authors frame their findings as input to guidelines and policy rather than as a verdict on one family. The review answers a question about populations. The question at the kitchen counter at a quarter past eleven is a question about one child, and it is a different question, resting on a thinner evidence base than the week's coverage implied.

The striking thing about this review is not what it found but how poorly it was reported. A pre-registered synthesis of 23 meta-analyses that publishes five null results alongside six positive ones, and whose largest precisely-estimated association accounts for about 2% of the variation between children, is a strong argument against the proposition that screens are the main force shaping a generation's development. It was covered as though it were the opposite. The honest summary is that screen use is one small, consistently measurable correlate of children's academic and emotional outcomes, sitting alongside much larger ones, and that after 2.8 million children the direction of the arrow remains genuinely unresolved. Anyone told this week that the science is now settled against screens was told something this paper does not say. A finding of this size belongs somewhere in a conversation about the shape of a childhood, but not at the centre of it.