Yes. `Overall_Impact` (Beneficial / Neutral / Negative) is the most natural grouping. The three groups differ sharply and consistently on almost every measure in the dataset (query results above). | Segment | Share of students | Daily hours | Sleep (hrs) | Stress | Mental health | GPA | Late-night use | |---|---|---|---|---|---|---|---| | Beneficial (3,681) | 81.8% | 4.42 | 7.03 | 11.16 | 84.26 | 3.55 | 53.5% | | Neutral (654) | 14.5% | 8.50 | 5.48 | 22.30 | 63.42 | 3.00 | 82.3% | | Negative (165) | 3.7% | 12.25 | 4.23 | 31.12 | 47.13 | 2.54 | 81.8% | - **Usage drives the ordering.** Average daily use rises from about 4.4 to 8.5 to 12.3 hours across the three groups. Sleep duration, sleep quality (2.72, 1.43, 1.04), mental health and GPA fall in step, while perceived stress rises. - **Late-night use separates the groups only partly.** About 82% of both the Neutral and Negative groups use social media late at night, against 53.5% of the Beneficial group. It splits Beneficial from the other two but does not distinguish Neutral from Negative. - **Weekend extra hours do not help.** They are nearly flat across groups (1.80, 1.79, 1.73), so weekend use is not a useful segmentation axis. - **The groups are very unequal in size.** The Negative group is only 165 students, so any further breakdown of it (by platform or age, for example) will rest on small counts. **Caveats** - `Overall_Impact` looks like an outcome or label, either self-reported or derived from the other measures. That would partly explain why the groups separate so cleanly. It is a good descriptive segmentation, but it is not evidence that usage causes the outcomes. The data is observational. - Because the groups track daily usage hours so closely, a usage-based banding (for example under 5, 5–9, and over 9 hours) would probably give a similar split without relying on the label. I did not test that. - I did not check whether the demographic and platform columns (age, gender, academic level, platform, device) add structure beyond this split. They are the natural next cuts, and pre-aggregated tables exist for them. ```chart {"kind":"bar","x":"Overall_Impact","series":["avg_daily_hours","avg_stress"],"title":"Usage hours and stress by overall impact"} ```