Screen Use Tracked Lower Language Scores, Not Reading or Math

It is a familiar scene. A parent at the kitchen counter near eleven at night, the dishes half done, scrolling the day's news on the same phone the children were told to put away at dinner. The headline says screens are linked to poorer development in 2.8 million children, and the question that follows it is not abstract at all. It is whether the tablet that bought forty minutes of quiet on Tuesday is quietly costing something, and whether that something turns up later in a child's vocabulary or on a report card.

The paper behind that headline was published in JAMA Pediatrics on 28 September 2026. It rewards a close reading, because what it found is narrower and stranger than the coverage suggests, and because the places where it found nothing are as informative as the places where it found something.

The paper is an umbrella review, which is a particular kind of summary: not a study of children, but a study of meta-analyses, each of which was itself a pooled summary of earlier studies. Al-Jbouri and colleagues searched Embase, MEDLINE and PsycINFO from each database's inception through 25 November 2025, and kept meta-analyses of population-based samples of people aged 18 years or younger published in English.

Twenty-three meta-analyses met the bar. Together they represented 2,817,252 children across 64 countries. Two reviewers independently extracted the data, at the level of each meta-analysis and at the level of the primary studies inside it, and converted every reported effect into a Pearson correlation, r, so that findings measured on different scales could be set side by side. The outcomes the authors set out to examine were academic functioning, cognitive skills, socioemotional development and mental health.

That conversion into a common currency is what makes a review like this readable, and it is also what makes it easy to over-read. A correlation of 0.15 is a statement about how two measurements move together across a very large group. It is not a dose, a threshold, or a forecast about a particular seven-year-old.

Two associations landed on the academic side. Screen-based media exposure tracked lower academic achievement, r = -0.15 (95% CI, -0.23 to -0.07), and lower language skills, r = -0.14 (95% CI, -0.18 to -0.10). The language estimate is the tighter of the two: its confidence interval is narrow, which means the pooled evidence agrees with itself reasonably well about roughly how large the association is.

On the mental health side, the review reported higher internalizing behavior, the inward-turning difficulties such as anxiety and low mood, at r = 0.10 (95% CI, 0.07 to 0.13) for general screen use and r = 0.09 (95% CI, 0.06 to 0.12) for social media use specifically. Externalizing behavior, the outward-turning kind, came in at r = 0.11 (95% CI, 0.08 to 0.14).

Two further estimates were much larger and much shakier: emotional problems at r = 0.30 (95% CI, 0.08 to 0.50) and peer problems at r = 0.28 (95% CI, 0.04 to 0.49). Those intervals are wide enough to span nearly the whole distance between a trivial association and a substantial one. A reader who keeps the 0.30 and leaves the interval behind has taken the least reliable number in the paper and treated it as the most alarming.

Several outcomes showed no clear association, and the list is specific enough to matter. Executive functioning, the planning-and-inhibition skills that a great deal of screen-time worry is aimed at, came in at r = 0.01 (95% CI, -0.09 to 0.11), an estimate sitting on zero. Prosociality was r = -0.22 (95% CI, -0.43 to 0.01), an interval that crosses zero and so cannot be called a finding.

The academic nulls are the most striking, because they sit directly beside the academic findings. Mathematics scores: r = -0.10 (95% CI, -0.23 to 0.03). Science: r = -0.03 (95% CI, -0.15 to 0.11). Reading: r = -0.02 (95% CI, -0.12 to 0.09). So a measure called language skills moved, and a measure called reading did not, inside the same review.

That pairing is the most interesting thing in the paper, and the review does not resolve it. The two constructs are related but not identical, and they are measured in different ways across the underlying literature. A synthesis pitched at this altitude cannot say whether the gap reflects a real difference in how screen use touches each one, or a difference in how reliably each was captured in the studies underneath.

The authors also report that screen type, the subtype of behavior problems, publication year and age each moderated some of these associations, meaning the size of the association varied along those lines. The abstract does not publish the per-type breakdown, so which screen types carried more of the association and which carried less is not something this reading can supply.

A correlation of -0.14 is small. It is also drawn from a pooled population of more than 2.8 million children, which makes it unlikely to be noise. Associations that faint can still matter at population scale while remaining invisible inside any individual household, and most of the public argument about screen time consists of holding one of those two facts and discarding the other.

What the review does well is narrow the claim. It does not support a broad story about screens damaging children's thinking: executive functioning, the most directly cognitive outcome in the set, showed nothing. It supports something more specific and more modest, with associations concentrated in language, in overall academic achievement, and in emotional and behavioral difficulties, most of them small and several of them imprecisely estimated.

The authors frame the work as material for the people who write guidelines rather than for families directly, describing its purpose as informing evidence-informed guidelines and policies issued by medical and governmental organizations. That is a fair account of what an umbrella review is good for. It summarizes a literature. It does not examine a child.

The honest summary here is not that screens harm children, but that screen use is faintly and unevenly associated with a few outcomes and not clearly associated with several others. This is a careful piece of work precisely because it reports its nulls at the same volume as its findings: reading, mathematics, science and executive functioning all came back empty, and those are four of the outcomes most often invoked when the subject comes up. Two numbers deserve the attention. The first is the language association, r = -0.14, small but tightly estimated, and the most defensible finding in the paper. The second is the pair of large, wobbly estimates for emotional and peer problems, r = 0.30 and r = 0.28, which are instructive mainly as a demonstration of how easily a wide confidence interval becomes a confident headline on the way to a reader. The review is addressed to the organizations that write screen-time guidance, and it reads like it. Its most useful contribution to a family is probably the negative space it clears: it shortens the list of things the current evidence actually supports worrying about.