**The most surprising finding is how clean and one-directional the data is, not any single reversal.** Daily social media use tracks nearly every wellbeing measure about as strongly as real survey data ever does, and it all points the same way. **Correlations across all 4,500 students** (steps 0 and 3): - Daily usage vs. mental health index: **-0.85** - Daily usage vs. sleep hours: **-0.72** - Daily usage vs. GPA: **-0.71** - Daily usage vs. sleep quality: **-0.68** - Daily usage vs. perceived stress: **+0.75** - Sleep hours vs. mental health: **+0.77** - Stress vs. mental health: **-0.77** - GPA vs. mental health: **+0.66** Correlations in the -0.85 to -0.7 range are unusually strong for self-reported behavioral and psychological measures. Real student surveys typically show weak to moderate links. Here, sleep, GPA, stress and mental health all move together with usage almost as if they were one underlying "dose". **The self-assessed impact label follows the same gradient** (step 1): | Overall impact | Students | Avg daily hours | Avg mental health | Avg GPA | Avg sleep | |---|---|---|---|---|---| | Beneficial | 3,681 | 4.4 | 84.3 | 3.55 | 7.0 | | Neutral | 654 | 8.5 | 63.4 | 3.00 | 5.5 | | Negative | 165 | 12.3 | 47.1 | 2.54 | 4.2 | The headline is that about 82% of students are labelled "Beneficial", and they are simply the light users. The "Negative" group averages about 12 hours a day, nearly three times the "Beneficial" group, with mental health scores about 37 points lower. **Some things that don't move at all** (steps 3 and 6): - Weekend extra hours vs. daily usage: **-0.01**. Heavy weekday users don't use more on weekends, and light users don't either, which is odd behaviorally. - Age vs. GPA: **+0.03**. **How to read this.** - The relationships are almost perfectly monotonic, with essentially no noise from age, platform or other factors. That is more tidy than organic data usually is, even though the dataset is catalogued as organic. Treat the effect sizes with caution. - The data is observational, so it can't show that usage causes the harm. Poor sleep or low mental health could equally drive heavier use. The correlations come from the single-row aggregate queries in steps 0 to 6 over `raw.Social_media_impact_on_life`. The impact-group breakdown is the Overall_Impact aggregate in step 1.