**The most surprising finding is how clean and strong the relationship is between social media use and wellbeing. Real-world survey data rarely looks like this.** **What the data shows (4,500 students)** - Daily usage hours correlate at **-0.85** with Mental_Health_Index. Real observational studies of social media and wellbeing usually report weak correlations, so this is unusually tight. - The same pattern holds across the other outcomes. Usage correlates at **-0.72** with sleep duration, **-0.71** with GPA, **-0.68** with sleep quality and **+0.75** with perceived stress. - The outcomes also track each other. Sleep correlates at +0.77 with mental health, stress at -0.77 and GPA at +0.66. Almost every measure moves together. - The `Overall_Impact` label shows the same gradient: | Overall_Impact | Students | Avg daily hours | Avg mental health | Avg GPA | Avg sleep (hrs) | Avg stress | |---|---|---|---|---|---|---| | Beneficial | 3,681 | 4.4 | 84.3 | 3.55 | 7.0 | 11.2 | | Neutral | 654 | 8.5 | 63.4 | 3.00 | 5.5 | 22.3 | | Negative | 165 | 12.3 | 47.1 | 2.54 | 4.2 | 31.1 | **The counterintuitive part** - **Most students are labelled "Beneficial".** That is 3,681 of 4,500 (about 82%), and their average use is already 4.4 hours a day. Only 165 students (about 4%) are labelled "Negative". - **Some variables show no relationship at all.** Weekend extra hours vs. daily usage is **-0.01**, and age vs. GPA is **0.03**. A heavy weekday user is no more or less likely to add weekend hours, and older students do no better academically. - **The effects are close to a straight line.** The strong correlations combined with the clean "Beneficial / Neutral / Negative" gradient look more like a designed or generated relationship than organic survey noise. **Caveats** - The dataset is observational. These correlations do not show that usage *causes* worse sleep, stress or GPA. Students who are already stressed or sleeping badly may also use social media more. - I did not run a test for whether the data is synthetic. This is an inference from the unusually high correlations, so I would treat any policy conclusions drawn from it with caution.