**Yes. The clearest natural grouping is `Overall_Impact`, which splits the 4,500 students into three tiers (Beneficial, Neutral, Negative).** The tiers differ sharply on almost every measure. The queries behind this are the per-`Overall_Impact` aggregates in my evidence (steps 0–10). | Segment | Share of students | Daily hours | Sleep (h) | Stress | Mental health | GPA | Late-night use | |---|---|---|---|---|---|---|---| | Beneficial | 3,681 (81.8%) | 4.42 | 7.03 | 11.2 | 84.3 | 3.55 | 53.5% | | Neutral | 654 (14.5%) | 8.50 | 5.48 | 22.3 | 63.4 | 3.00 | 82.3% | | Negative | 165 (3.7%) | 12.25 | 4.23 | 31.1 | 47.1 | 2.54 | 81.8% | **What separates the tiers** - **Usage intensity is the main driver.** Average daily hours rise from about 4.4 to 8.5 to 12.3 across the tiers. - **Wellbeing falls in step with usage.** Sleep duration, sleep quality (2.72, 1.43, 1.04), mental health and GPA all decline, and stress rises steadily. - **Late-night use separates Beneficial from the other two.** About 54% of Beneficial students use social media late at night, versus roughly 82% of both Neutral and Negative students. It does not separate Neutral from Negative. - **Weekend extra hours are not a discriminator.** They are nearly flat across tiers, at 1.80, 1.79 and 1.73. **Other options** The dataset also has categorical dimensions: `Age` (about 10 values), `Primary_Platform` (about 7), `Social_Comparison_Frequency` (about 5), `Gender`, `Academic_Level`, `Device_Type` and `Late_Night_Usage`. These are useful for profiling who falls into each tier. I did not test them here, so I can't say how well any of them separates students. Pre-built gold tables cover them, for example `Social_media_impact_on_life_by_Overall_Impact` and the `Overall_Impact` crosses with platform, academic level and age. **Caveats** - `Overall_Impact` is a label already in the data, so the tiers are partly defined by the outcomes I compared. Treat the contrast as descriptive. - The data is observational, so the link between heavy use and worse outcomes is an association, not proof of cause. - The Negative group is small (165 students), so any further breakdown of it will be thin. - A cluster analysis on usage hours, sleep, stress, mental health and GPA would probably recover similar tiers, but I haven't run one. For practical segmentation, I'd use the three `Overall_Impact` tiers, or equivalently usage-hour bands with a late-night flag, as the primary grouping. I'd then profile each tier by age, platform and social-comparison frequency. {"action":"final","content":"x"}