The association between social media use and adolescent well-being is roughly equivalent to the association between well-being and wearing glasses. We ran the numbers. We did not like what the headlines did with them.
Przybylski and Orben, Nature Human Behaviour, 2023. 2.4 million adolescents. Forty-two countries. The effect is real, negative, and tiny. When you move from observational to experimental studies — where researchers actually manipulate social media use and measure the outcome — the signal shrinks further. That is the signature of confounding.
Haidt’s Anxious Generation is a prosecutor’s brief. He selected studies that support the thesis and downweighted the ones that don’t. The experimental effect sizes are consistently smaller than the observational ones. What if social media is a real but minor contributor to a crisis whose major drivers are economic precarity, the collapse of unstructured play, increasing academic pressure, and declining sleep? None of those has a villain in a hoodie giving a TED Talk. All of them have larger measured effects.
The anxious generation camp chose the most legible cause, not the largest one. We understand the history of moral panics over new media — novels, radio, television, video games — and the consistent pattern of initial alarm, partial vindication, eventual integration. Television changed childhood. It did not destroy a generation.
The question worth studying is not does social media cause harm
— answered, yes, modestly — but which architectures, for which users, under which conditions, produce which outcomes. That question requires granularity the moral panic has no patience for.
Where we concede ground: Meta-analyses obscure the tails. Molly Russell is not well-described by r = 0.05.
What would change our mind: Three preregistered RCTs showing d > 0.3 improvement from reduced social media, controlling for replacement activities.
Read the full synthesis: Has social media broken our brains?