“Panic First, Evidence Later”? Surely That’s Not How We Do It in the EU!

At a roundtable in early June that brought together children’s rights experts, child safety researchers, and policymakers, not a single researcher present expressed support for a social media age ban. When a policymaker announced his country was about to introduce one on the strength of the scientific evidence, the sighs from the researchers in the room were audible. The gap between what the science shows and what is being legislated in its name has grown wide enough to hear.

“Panic First, Evidence Later”, a June 2026 systematic review by Sara M. Grimes and Ujunwa Ohakpougwu at McGill University’s Kids Play Tech Lab, documents that gap in detail. It draws on six peer-reviewed domain experts: developmental psychologists, media researchers, and children’s rights scholars who have spent careers studying the questions Jonathan Haidt addresses in The Anxious Generation, which he now describes as a “movement”. Their critiques converge: the causal claims driving current legislation lack evidential support, the methodology behind the most-cited findings is systematically flawed, and the policies being enacted are poorly calibrated to the actual distribution of risk. These are the same findings we have collected in our information labs Social Media Ban Repository, and highlighted in our previous blog posts.

The Expertise Gap: Vote Counting Is Not Research

A prior question sits underneath all of this: whose evidence counts, and why? Haidt holds a chair at NYU Stern built on moral psychology, moral emotions, and political psychology. Developmental science, media research, and child psychology fall outside his research area. His published work on social media and adolescent mental health consists of reviews of existing research, not original empirical studies.

His method, compiling studies pointing in one direction and treating their accumulation as proof of a strong effect, is what critics call “vote counting”: tallying studies regardless of methodological quality, sample size, or participant age. A handful of large, well-designed studies finding no effect carry more evidential weight than many small, methodologically weak studies finding harm. That distinction is not something vote counting can capture.

The report also flags a more basic problem. Haidt’s policy prescriptions target children and teenagers. A review of every experimental study in his self-curated evidence document shows the overwhelming majority were conducted on college students and adults aged 18 to 35. Only one study involved actual adolescents. Policymakers citing the book as their scientific basis for restricting children are drawing on literature that barely looked at them. And probably did not speak with them.

There is an irony worth registering. Haidt’s own previous books warned against exactly this kind of reasoning. In The Righteous Mind he showed how people deploy evidence to justify intuitions already held. In The Coddling of the American Mind he warned that protecting children by removing everything potentially risky backfires, producing more anxiety, not less. His critics argue he is now doing both simultaneously.

Correlation and Its Limits: Treating a Symptom as a Cause

The most fundamental scientific challenge comes from Candice Odgers, Associate Dean for Research at UC Irvine and one of the foremost authorities on child and adolescent mental health risk. Her 2024 review in Nature makes the point plainly: the book’s core claim is not supported by the science.

The evidence Haidt draws on is overwhelmingly correlational. Two things happening at the same time cannot establish that one causes the other. In this case, the causal arrow may run the wrong way entirely: young people who are already struggling may seek out social media more, rather than social media producing the struggle. The Adolescent Brain Cognitive Development study, the largest long-term study of adolescent brain development in the United States, found no evidence of drastic changes linked to digital technology use. A 72-country analysis, by Vuorre and Przybylski, published in Royal Society Open Science,  found no consistent global association between social media penetration and changes in well-being.

Odgers’ summary is blunt: hundreds of researchers have searched for the large effects Haidt describes and have not found them. When longitudinal associations are found, they suggest that young people with existing mental health problems use social media more, not that social media is producing those problems. The policy is treating a symptom as a cause.

What the Effect Sizes Show

Amy Orben and Andrew Przybylski, working across Cambridge and Oxford, have produced some of the most technically rigorous large-scale studies of screen time and adolescent well-being ever conducted, analysing data from 355,000 participants. Their finding: digital technology use explains at most 0.4% of the variation in adolescent well-being. Wearing glasses and eating potatoes showed similarly sized associations in the same datasets.

That comparison is not a joke. It is a methodological point about the scale of the claimed effect relative to the scale of the claimed crisis. An association as weak as that between potato consumption and banning a communications medium for an entire age group does not justify it.

The gap between that finding and Haidt’s framing is methodological. Orben and Przybylski applied specification curve analysis, which tests all defensible analytical approaches to a dataset simultaneously rather than selecting for the strongest signal. Studies that use analytical flexibility, choosing the approach most likely to produce a significant result, inflate effect sizes systematically. Vote counting does the same. Neither is a reliable basis for major legislation.

The methodology behind the measurement exacerbates the problem. Participants consistently over- and under-estimate their social media usage. Device-level log data produces substantially weaker associations than self-reports. The evidence base Haidt draws on is disproportionately built on self-reported screen time, which the research literature has demonstrated to be systematically unreliable.

72 Countries, No Consistent Pattern

One of Haidt’s more rhetorically powerful arguments is the international scope of the problem. If smartphones and social media are causing the crisis, the same pattern should appear wherever they spread. The predicted pattern does not appear, as evidenced by the 72-country analysis by Vuorre and Przybylski, the most geographically comprehensive study of this question conducted to date.

Critics have observed that Haidt draws heavily on American data, the country with the most marked trends, while citing international comparisons selectively. The patterns most prominent in his analysis may reflect domestic American conditions: the opioid crisis, racial inequality, gun violence, school safety culture, and healthcare access, rather than a universal effect of the phone-based childhood. European countries with comparable smartphone penetration show no comparable youth suicide trends. Canada’s national youth suicide rate declined 24% between 1981 and 2017. A universal cause should produce more consistent patterns than this.

Effects That Vary by Person

Setting aside causation and scale, there is a deeper problem with how any effects are distributed. Patti Valkenburg, distinguished university Professor and founding director of the Center for Research on Children, Adolescents, and the Media at the University of Amsterdam, has developed within-person research designs tracking how the same individual’s social media use relates to their own well-being over time. What she finds cuts against the logic of uniform age-based prescriptions.

Social media effects are highly variable across individuals. In a series of person-specific studies, the “passive social media use harms well-being” hypothesis was rejected for 80% of adolescents in within-person designs. Most young people show little to no relationship between social media use and well-being. A minority show negative effects. Some show positive ones.

The studies Haidt draws on are predominantly cross-sectional, comparing heavy users to light users at one point in time, which consistently overstates associations relative to within-person designs. Outcomes depend on context, pre-existing vulnerability, and the quality of online relationships. Restricting access uniformly to protect a minority already vulnerable for other reasons, while removing a resource the majority use without harm, is not a calibrated response.

Bans Punish Children for Platform Failures

Sonia Livingstone, Professor of Social Psychology at LSE and one of the world’s foremost authorities on children’s digital lives, clearly describes the real-world implications of a ban. Blanket age-based restrictions remove a resource that benefits many young people in order to protect a vulnerable minority whose greatest risks derive, in most cases, from pre-existing conditions rather than the platforms themselves.

Age bans collapse protection rights and participation rights entirely on the side of protection, ignoring the genuine benefits of digital participation for LGBTQ+ young people, those with disabilities, and those in isolated communities whose primary support is online.

As Livingstone puts it, the argument for a ban is an admission of failure to regulate platforms, resolved by restricting children instead. Age bans do not touch algorithmic design, data collection practices on minors, or the structural conditions — inequality, family disruption, underfunded mental health services — where the evidence of causal contribution is considerably stronger.

The Recurring Panic Pattern

Christopher Ferguson, a psychologist at Stetson University, places the current debate within a documented historical sequence. Comic books caused juvenile delinquency in the 1950s. Rock music corrupted youth in the 1960s. Dungeons and Dragons led teenagers into occult murder in the 1980s. Violent video games produced aggression in the 1990s. In each case, a small number of persuasive voices drove public alarm; in each case the science failed to support it; in each case the panic subsided and the regulations with it.

Ferguson was one of the researchers Haidt consulted before publication, and his advice was not to publish. His 2025 peer-reviewed analysis finds that youth suicide patterns correlate with multiple environmental factors, including income inequality, family structure, and adult suicide trends. Attributing those patterns to a single cause risks what he calls ecological fallacy: mistaking correlation with a simultaneous factor for a causal mechanism. Haidt’s trend graphs do not control for any of those confounds.

Evidence Should Now Prevail: We Owe Teens That Much

The researchers who sighed in that roundtable room were not indifferent to children’s safety. Quite the opposite in fact. They were expressing their frustration at what qualifies as “science” these days in policy circles.

“Panic First, Evidence Later” does not argue that social media carries zero risk or that youth mental health requires no response. The legislation currently being enacted is disproportionate to the evidence and crowds out attention from structural factors where causal contribution is considerably stronger: inequality, family disruption, educational pressure, climate anxiety, deteriorating mental health infrastructure. Platform design accountability, age-appropriate design codes, genuine enforcement of existing protections, and targeted support for the young people actually at risk are all available instruments. Several already exist in European law and go chronically underenforced.

None of that requires drawing a random age line and calling it science.