Growing up online

A disagreement between researchers reading the same evidence

A The claim that smartphones and social media have damaged a generation's mental health is now made in legislatures, school policies and best-selling books, and is treated in public discussion as established. Among researchers it is contested, and the contest is unusually instructive, because both sides are looking at the same data and disagreeing about what can be concluded from it rather than about what it says.

B It is worth separating the several claims usually bundled together. One is that adolescent mental health has deteriorated; another that the deterioration was caused by these products; a third that particular remedies would reverse it. The first is very well supported, the second is disputed, and the third has barely been tested at all. Public discussion routinely treats agreement on the first as though it settled the other two. The pattern that prompted the alarm is real. In several wealthy countries, measures of adolescent depression, anxiety, self-harm and suicide rose from around 2012, after a long period of stability or decline; the rise was steeper among girls; and the timing coincides with the point at which smartphone ownership and social media use became close to universal among teenagers. Hospital admissions for self-harm, which do not depend on anyone's self-report, moved in the same direction, which disposes of the simplest objection — that young people had merely become more willing to describe distress.

C The case against the causal reading begins with what the correlations look like when measured directly. Studies relating individual screen time to individual wellbeing find associations that are consistently negative and consistently tiny, of a size comparable to the association between wellbeing and eating potatoes — a comparison introduced by one research group to make exactly this point. If the technology were doing what is claimed, the effect ought to appear in these measurements, and it does not appear at anything like the required magnitude. The response that these studies measure the wrong thing is reasonable and cuts both ways, since a measure too crude to detect a large effect is also too crude to have generated the alarm in the first place.

D Several defences of the causal claim have been offered against that objection. Average screen time may be the wrong measure, since the same three hours spent scrolling through images of unattainable bodies and spent messaging close friends are unlikely to have the same effect. The harm may be collective rather than individual: if the arrival of these platforms changed the social environment of an entire cohort — how friendships work, where status is contested, what is happening at three in the morning — then comparing heavier and lighter users within that cohort would understate it, because everyone is affected by a norm regardless of their own use. This is a serious argument and it is also, awkwardly, one that predicts the absence of the evidence that would confirm it.

E The experimental evidence is thinner than either camp acknowledges. Studies that pay people to deactivate their accounts for a period generally report modest improvements in reported wellbeing, but they are conducted mostly on adults, run for weeks rather than years, and cannot reproduce the situation of a teenager whose friends are all still using the platform. School phone bans have produced mixed results, with the clearest benefits in attention and attainment rather than in mental health, and the studies with the strongest designs tend to report the smallest effects.

F Underlying the disagreement is a measurement problem that is rarely stated plainly. Self-reported screen time correlates poorly with actual logged use; the platforms hold the data that would settle several of these questions and do not release it; and the mental-health measures used before and after the critical period were not designed to be compared with each other. A question of this importance is being argued from instruments too blunt to answer it, which is why intelligent people can examine the same literature and reach opposite conclusions in good faith.

G The practical position does not depend on resolving this. Several of the measures proposed — restricting phones in schools, delaying the age at which children are given unrestricted accounts, requiring platforms to turn off features designed to maximise the time a child spends on them — are justified by considerations that stand independently of the mental-health question, including attention, sleep and the ordinary principle that a product should not be engineered to be difficult for a child to stop using. It is possible to think the evidence for a mental-health catastrophe is weaker than advertised and still think that a fourteen-year-old should not be served an infinite feed at midnight. Insisting that the case rests on the strongest version of the claim is a mistake, because the strongest version may not survive, and the reasonable measures should not fall with it.