What the Research Actually Says About Social Media and Teenagers (It's Not What Either Side Claims)

Jonathan Haidt's book The Anxious Generation has had the kind of cultural impact that very few academic-adjacent works achieve. It has become the reference text for a school of parental concern that I hear articulated in some version at almost every parent information evening: smartphones and social media are responsible for a mental health crisis in adolescents, and the solution is strict limits, ideally legislated ones.

I've read the book. I've also read a fair amount of what the critics have written. And I want to try to give you a genuinely honest account of where the evidence sits — not because I think Haidt is a crank, but because a community making policy decisions (including the individual household decisions that aggregate into culture) should be working from the actual state of the evidence rather than a confident simplification of it.

The Haidt argument

The core claim is this: adolescent mental health in Western countries began declining sharply around 2012, which corresponds closely with the widespread adoption of smartphones and the shift of teenage social life onto social media platforms. Girls, in particular, show elevated rates of anxiety, depression, and self-harm. Haidt argues that social comparison on visual platforms, cyberbullying, sleep disruption, and the displacement of in-person interaction are the causal mechanisms. The solution he advocates is structural: later phone adoption, age restrictions on social media platforms, more unsupervised physical play.

The correlation is real. The mental health decline is documented across multiple datasets. The timing is striking. And Haidt's proposed mechanisms are plausible — there are plausible reasons why constant social comparison at the most identity-sensitive period of development could cause harm.

I do not dismiss this argument. I think it is raising something real.

The complications

Here is where I ask you to stay with me, because this requires holding more than one idea at once.

The pre-dating problem. When researchers look closely at the timeline in some datasets — particularly in the UK — the mental health decline begins slightly before the smartphone adoption curve. This doesn't disprove causation, but it complicates it. If the crisis began before the cause, you need either a different cause or a different timeline.

The gender asymmetry. The mental health effects are substantially larger for girls than for boys, which Haidt acknowledges. But this asymmetry is hard to fully explain by social media exposure alone, since boys use social media too. There are competing hypotheses: visual social comparison platforms are more central to girls' use patterns; pre-existing anxiety and depression (which are more prevalent in girls) are amplified by social media; the offline social dynamics being mediated by the platforms differ by gender. The truth is probably some of all of this, but "social media" as a uniform cause struggles to explain the disparity.

Correlation vs. causation (the familiar but important problem). Several large-scale studies — including work using UK Millennium Cohort data by Orben and Przybylski — find that the effect of social media on adolescent wellbeing, while detectable, is very small in population terms. The effect size is comparable to the effect of eating potatoes or wearing glasses. Haidt disputes both the methods and the interpretation of these studies. The dispute is ongoing and technically complex. What I take from it is that the effect is real but the confidence about its magnitude is lower than the policy debate implies.

Platform differences matter enormously. Instagram and BeReal are not the same thing. Discord and TikTok are not the same thing. Using "social media" as a single category in research papers — and in policy — blurs distinctions that may be crucial. The visual social comparison dynamics of Instagram are different from the group-chat communication dynamics of WhatsApp. We are not seeing a uniform effect across platforms.

Pre-existing conditions. Children who are already anxious or depressed use social media differently from children who are not, and show larger negative effects. This doesn't mean causation runs only one way — social media can clearly worsen pre-existing anxiety. But it does mean that studies without adequate controls for prior mental health are going to overestimate effects.

The Hong Kong data specifically

HK data adds some specific complications to the global picture.

The mental health pressures on HK adolescents are not principally driven by social media dynamics. They include among the most competitive academic environments in the world, high-stakes exam culture, concerns about the city's political direction that have been acutely present for young people, family financial pressure in one of the world's most expensive housing markets, and, for many students, commuting times that leave very little discretionary time for anything.

Any study of social media effects in HK that doesn't control for these contextual factors is measuring something confounded. When I look at the anxious students I know, their anxiety is usually connected to exams, family pressure, and uncertainty about their future. Social media amplifies and mediates those anxieties. It is not usually the root cause.

I also note that HK's 43% of primary and 81% of secondary students exceeding the two-hour daily screen time recommendation is a population-level figure that includes very different kinds of use. Watching Hong Kong drama with a grandparent, communicating with a parent who is commuting home, playing a game with school friends, and doomscrolling social comparison content at midnight are all "screen time." They are not the same intervention target.

My conclusion

The research supports concern and warrants action. It does not support the specific confidence that legislative bans are the right lever, or that reducing screen time alone (without attending to what replaces it) will produce the hoped-for outcomes.

What the evidence does support: platform design regulation to reduce engagement-maximisation features that have documented harms. Later school start times (sleep is more consistently linked to adolescent mental health than screen time, and phones at night are partly a sleep issue). Genuine curricular space for social-emotional learning, including digital literacy. And consistent, non-panicked parental engagement with what children are actually doing online.

The ban advocates are not wrong to be worried. They are wrong to be so certain about the remedy. The counter-advocates who dismiss the concern entirely are reading selectively. The honest position is somewhere in the middle, which is always less satisfying but more useful.