The moral panic surrounding adolescent social media use has driven sweeping policy debates, yet most supporting evidence comes from cross-sectional snapshots rather than genuine long-term follow-up. A rigorous longitudinal design tracking the same individuals across five years now offers a notably different picture — one that should prompt researchers and commentators to revisit confident causal claims.
Drawing on a nationally representative Swedish cohort, investigators surveyed 5,535 ninth-graders (ages 15–16) in 2017 and recontacted 3,193 of them in 2022. Daily social media hours were self-reported at baseline; anxiety and depression were assessed at follow-up using the validated Patient Health Questionnaire-4. In unadjusted models, early social media exposure did correlate with higher symptom scores five years later — but once confounders were entered, both associations collapsed to near-zero (depression: β = 0.02, 95% CI −0.00 to 0.05; anxiety: β = −0.00, 95% CI −0.03 to 0.03). Subgroup analyses found no meaningful effect modification by gender, socioeconomic status, or physical activity level.
This finding sits within a contested literature. Several high-profile studies — particularly those using pre-teen and early-teen samples from the United States and United Kingdom — have reported modest negative associations between heavy platform use and well-being, especially for girls. Yet effect sizes across that literature are frequently small, and replication analyses of large datasets (including those by Amy Orben and Andrew Przybylski) have repeatedly found that screen time explains only a marginal share of variance in adolescent mental health, comparable to wearing glasses or eating potatoes. The Swedish result aligns with that skeptical strand of evidence and raises the possibility that unmeasured third variables — pre-existing vulnerability, family dynamics, or peer environment — drive both heavy social media use and later distress.
Key limitations include substantial attrition (42% of the original cohort was lost to follow-up), reliance on self-reported usage, and the inability to capture platform-specific or content-specific effects that may matter more than raw hours. This study is best read as an important corrective, not a definitive exoneration.
