In 2023, the American Psychological Association published guidance advising adolescents to limit social media use and calling for parental supervision of content. In 2024, a large pre-registered study in Nature Human Behaviour found no consistent relationship between social media use and adolescent wellbeing. Both drew on peer-reviewed research. The contradiction is not unusual. It is the normal state of this particular literature.

Here is what the science has actually established with reasonable confidence. First: passive consumption is more consistently associated with poor outcomes than active use. Scrolling without interaction, watching other peoples lives accumulate without participating, is the behaviour that correlates most reliably with lowered mood, particularly in girls. Active communication, creating content, and using platforms to organise social activity shows weaker or negligible negative effects in most datasets.

Second: timing matters more than duration. Late-night use that displaces sleep is the mechanism behind much of what looks like a screen time effect in the data. The association between heavy social media use and anxiety frequently disappears when sleep quality is controlled for. This is not a trivial finding. It relocates the intervention. The question is not how many hours a day, but when and at the cost of what.

Third: the most vulnerable adolescents are most affected. Young people with pre-existing anxiety, depression, or low social connection show stronger negative associations with problematic use. The research is not finding a population-wide harm. It is finding a concentrated harm in already-struggling young people. This changes the risk calculus for parents significantly.

Fourth: effect sizes are consistently small. The 2022 analysis by Andrew Przybylski and Amy Orben, which examined data from over 350,000 adolescents, found that the association between technology use and wellbeing was comparable in magnitude to the effect of wearing glasses or eating potatoes. Statistically detectable. Not clinically meaningful at a population level.

None of this means screens are neutral. It means the conversation has been substantially louder than the evidence justifies. Specific behaviours, specific populations, specific contexts: that is where the real signal is. Blanket limits applied without that specificity are unlikely to do much.