Three questions stand out. The overall correlations I computed over all 4,500 students show what is worth digging into. I haven't yet run the segment-level cuts, so those are proposed questions, not findings. **What the data shows so far** - Daily usage averages 5.3 hours and average GPA is 3.43. - More usage goes with worse outcomes: - Mental health index: r = -0.85 - Perceived stress: r = +0.754 - Sleep duration: r = -0.716 - GPA: r = -0.71 - Better sleep goes with better outcomes: - Sleep duration and GPA: r = 0.595 - Sleep quality and mental health: r = 0.662 **1. Is there a usage threshold where wellbeing starts to fall sharply?** The overall correlations are strong, but a single r hides the shape. If mental health, stress, sleep and GPA hold steady up to some number of hours and then drop, that threshold is a concrete limit to recommend. If the decline is gradual, a blanket cap is harder to justify. Banding `Daily_Usage_Hours` (and adding `Weekend_Extra_Hours`) against the five outcome measures would answer this. It directly serves the decision on whether to recommend daily-use limits. **2. How much of usage's effect on GPA runs through sleep?** Usage is linked to lower GPA (-0.71) and to shorter sleep (-0.716), and sleep is linked to higher GPA (0.595). So sleep may be the route through which heavy use hurts grades, or usage may hurt grades independently. Comparing the usage–GPA relationship within sleep bands, and splitting by `Late_Night_Usage`, would separate the two. That tells you whether to target late-night use and sleep, or total screen time. **3. Which segments are hit hardest once usage is held constant?** Platform, device type, academic level, gender, age, and `Social_Comparison_Frequency` may differ in risk simply because those groups use more. Comparing mental health and stress across these groups within the same usage band shows who is more vulnerable per hour of use. The dataset already has pre-aggregated tables for each of these cuts. This is the question that tells outreach teams which groups to prioritize, and whether to target social comparison behavior rather than time online. **Caveats** - The data is observational, so all of this shows association, not causation. Heavy use could be a symptom of poor mental health, not just a cause. - A correlation of -0.85 between usage and mental health is unusually strong for survey-type data. Check the data's provenance, and look for non-linear patterns or outliers, before treating it as a real-world effect size.