PISA 2022 — the world's largest education assessment. The relationship between internet use and teen wellbeing is not an American artifact. It replicates across every cultural context we tested.
Paper 166 showed which features predict harm using U.S. data. This paper asks: does it replicate? We used OECD's PISA 2022 — 613,744 students in 80 countries, the same test given worldwide.
Among the 12 Western European nations, the correlation between internet use and life satisfaction is strong, negative, and statistically significant — even after controlling for GDP.
These 12 countries share similar healthcare, education, and social support systems. GDP variation is small. Cultural differences exist but the fundamental conditions of adolescence are comparable. This makes it the cleanest natural experiment in the dataset.
The correlation survives controlling for GDP. Wealthy countries are not simply sadder. It is the internet use variable that carries the signal.
The global ecological correlation is weaker — we report that honestly. Different regions have different confounders: internet infrastructure, survey response patterns, cultural attitudes toward mental health disclosure.
But the gender disparity replicates almost everywhere. In 91% of the 47 countries with gendered data, girls show larger negative effects. That consistency across radically different cultures demands explanation.
5.6 times more affected than boys. Not in one country. In almost all of them. Across cultures that agree on almost nothing else.
p < 0.000001 — the probability this is random is less than one in a million.
The features that score highest in Paper 166 — opaque recommendation, social comparison, beauty filters — are the features that differentially target appearance-based social evaluation. Platforms did not intend to harm girls. The geometry of the features does it automatically.
PISA includes individual-level internet use data. The dose-response relationship (more hours, lower satisfaction) is statistically significant at p = 0.007. This is not just a country-level ecological pattern. It holds within individuals.
The social media debate has been dominated by U.S. data. Paper 166 used CDC YRBS (U.S. only). This paper proves the pattern is not cultural — it is structural.
Opacity features (algorithmic feeds, opaque recommendations, autoplay) dominate the signal across all regions. The specific platforms differ — TikTok dominates in Asia, Instagram in Europe, YouTube everywhere — but the features that predict harm are the same features.
If the pattern were cultural, regulation would be a local question. It is not. The same design features produce the same outcomes on every continent. This is a design problem with a design solution: change the features, change the outcome.
Two papers. Two datasets. One pattern. The features are verifiable. The harm is measurable. The fixes are specific.