**Metric: average sleep latency (`sleep_latency_min`). Category: `primary_bedtime_app`.** This is a judgement call, not a computed ranking. I chose them because the dataset's purpose is to decide whether bedtime phone habits drive poor sleep, and the app is the one behavioural segment that clearly separates outcomes. **How sleep latency varies by app** (from the per-app breakdown I queried): | App | Avg latency (min) | Users | |---|---|---| | TikTok / Reels | 46.1 | 2,198 | | Instagram / Reddit | 42.5 | 1,630 | | YouTube | 39.1 | 1,928 | | Messaging / Chat | 38.3 | 847 | | Streaming (Netflix/Hulu) | 36.4 | 1,301 | | News / Reading | 33.2 | 596 | - **Size of the gap:** TikTok / Reels users take about 13 minutes longer to fall asleep than News / Reading users (46.1 vs 33.2), roughly 39% more. The medians show the same order (43.0 vs 31.8). - **Other outcomes move the same way, but less:** - Next-day fatigue averages 4.16 for TikTok / Reels and 3.33 for News / Reading. Instagram / Reddit is next highest at 3.94. - Average sleep hours barely differ, from 6.17 (TikTok / Reels) to 6.46 on the high end (News / Reading median). - Latency is therefore the clearest differentiator between apps. - **Behaviours that don't differ much by app:** average screen brightness is about 55% for every app. Post-5pm caffeine (about 32–35 mg) and physical activity (about 35–36 min) are also close across apps. **Caveats** - The data is observational, so this shows association, not cause. - The apps also differ in who uses them. TikTok / Reels users are youngest (average age 31.1) and News / Reading users oldest (38.0). Age could account for part of the latency gap. - I didn't test whether other categories, such as chronotype or occupation, separate latency more strongly. ```json {"charts":[{"kind":"bar","x":"primary_bedtime_app","series":["sleep_latency_min__avg"],"title":"Average sleep latency (min) by primary bedtime app"}]} ``` {"action":"final","content":"**Metric: average sleep latency (`sleep_latency_min`). Category: `primary_bedtime_app`.** This is a judgement call, not a computed ranking. I chose them because the dataset's purpose is to decide whether bedtime phone habits drive poor sleep, and the app is the one behavioural segment that clearly separates outcomes.\n\n**How sleep latency varies by app** (from the per-app breakdown I queried; the table is rendered above):\n\n- **TikTok / Reels:** 46.1 min average (2,198 users)\n- **Instagram / Reddit:** 42.5 min (1,630 users)\n- **YouTube:** 39.1 min (1,928 users)\n- **Messaging / Chat:** 38.3 min (847 users)\n- **Streaming (Netflix/Hulu):** 36.4 min (1,301 users)\n- **News / Reading:** 33.2 min (596 users)\n\n- **Size of the gap:** TikTok / Reels users take about 13 minutes longer to fall asleep than News / Reading users (46.1 vs 33.2), roughly 39% more. The medians show the same order (43.0 vs 31.8).\n- **Other outcomes move the same way, but less:**\n - Next-day fatigue averages 4.16 for TikTok / Reels and 3.33 for News / Reading. Instagram / Reddit is next highest at 3.94.\n - Average sleep hours barely differ, from 6.17 (TikTok / Reels) to about 6.38 (News / Reading).\n - Latency is therefore the clearest differentiator between apps.\n- **Behaviours that don't differ much by app:** average screen brightness is about 55% for every app. Post-5pm caffeine (about 32–35 mg) and physical activity (about 35–36 min) are also close across apps.\n\n**Caveats**\n- The data is observational, so this shows association, not cause.\n- The apps also differ in who uses them. TikTok / Reels users are youngest (average age 31.1) and News / Reading users oldest (38.0). Age could account for part of the latency gap.\n- I didn't test whether other categories, such as chronotype or occupation, separate latency more strongly.","charts":[{"kind":"bar","x":"primary_bedtime_app","series":["sleep_latency_min__avg"],"title":"Average sleep latency (min) by primary bedtime app"}]}