Here are the three questions I'd pick, each backed by the correlation and filter profile I ran over all 8,500 users. The data is observational, so every result below is an association, not proof of cause. ## 1. How much does bedtime phone use drive sleep latency, and does it carry through to sleep debt and fatigue? - Bedtime phone minutes and sleep latency correlate at **0.86**, the strongest behavioural link in the data. - Phone minutes and total sleep hours correlate at **-0.55**. - Latency and next-day fatigue correlate at **0.74**. - Sleep hours and fatigue correlate at **-0.88**, the strongest of all. - This suggests a chain: phone use, then slower sleep onset, then less sleep and more sleep debt, then more fatigue. - Follow-ups worth running: a dose-response curve in phone-minute bands, and a check of whether the effect holds across chronotype, occupation and bedtime app. - A correlation of 0.86 is unusually tight for real behavioural data, so I'd also check whether the data is cleaner than expected. ## 2. Do the commonly advised fixes (blue-light filters, lower brightness, a caffeine cutoff, more exercise) actually matter? - **Blue-light filter:** average latency is **39.6 min with the filter on** versus **41.6 min with it off**. That is a gap of about 2 minutes, small next to the phone-minutes effect. - **Brightness:** the correlation with latency is only **0.09**. - **Physical activity:** the correlation with latency is about **-0.01**, effectively none. - **Caffeine after 5pm:** the correlation with total sleep is **-0.04**, but with deep-sleep percentage it is **-0.34**. Caffeine may hurt sleep quality more than sleep duration. - The question to answer is which lever has real leverage. The early reading is that cutting phone minutes matters far more than filters or brightness, and that caffeine affects deep sleep rather than total hours. - A follow-up is to compare filter effects after holding phone minutes constant. The filter gap may just reflect different phone habits between filter users and non-users. ## 3. Which segments are most at risk, and where should a sleep-hygiene intervention be targeted? - The data has classifiers for chronotype, occupation type, primary bedtime app, gender and sleep-debt category. - Pre-aggregated gold tables already cover these cuts, for example by chronotype and bedtime app, or by occupation and sleep-debt category. - The question is which combinations (for example, evening chronotypes who use a particular app) have the highest latency, the worst sleep debt and the highest fatigue. - It also asks whether those groups differ because they use their phones more or because the same phone use hits them harder. That decides whether to target heavy users or particular groups. ## Caveats - `bedtime_phone_minutes` is flagged as person-identifying, so I would only report it in aggregate or in bands. - `sleep_debt_category` is derived from `total_sleep_hours`. Using it as an outcome alongside sleep hours would double-count the same signal.