**Yes. `Overall_Impact` (Beneficial / Neutral / Negative) is the most natural segmentation, and it lines up with a clear usage-intensity gradient.** The rendered result tables show three groups that differ sharply on nearly every measure: | Segment | Share of students | Avg daily hours | Avg sleep (hrs) | Avg stress | Avg mental health | Avg 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% | - **Gradient:** Moving from Beneficial to Neutral to Negative, daily usage rises (4.4 to 8.5 to 12.3 hours). Sleep, sleep quality, mental health and GPA fall, and stress rises. The groups are well separated rather than overlapping. Sleep quality drops from 2.72 to 1.43 to 1.04. - **Late-night use:** It is much higher in the Neutral and Negative groups (about 82%) than in the Beneficial group (53.5%). The two worse groups look alike on this measure, so it doesn't separate them from each other. - **Weekend extra hours:** These barely differ across groups (1.80, 1.79, 1.73). Weekday daily usage, not weekend extra, drives the separation. - **Size:** The groups are very unbalanced. Negative is only 165 students, so any sub-breakdown within it will be thin. **Caveats and other options** - `Overall_Impact` looks like an outcome label, probably self-reported or derived from the other measures. That would partly explain why it separates the groups so cleanly. The data is observational, so this shows association, not cause. - If you want segments that don't depend on that label, the next-best candidates are a **daily-usage band** (e.g. under 5, 5–9, and over 9 hours) combined with **late-night use**. These are the behavioural drivers that the three groups differ on. - Demographic splits (`Age`, `Gender`, `Academic_Level`) and context splits (`Primary_Platform`, `Device_Type`, `Social_Comparison_Frequency`) are available as pre-aggregated tables. I haven't checked whether they separate students as strongly as the impact groups, so I can't say whether they are useful segments. The segment profile comes from the `raw.Social_media_impact_on_life` aggregation grouped by `Overall_Impact` (results in steps 4–5).