## The 3 most interesting questions this dataset can answer The dataset has 4,500 students, one row each. It covers social media use, sleep, stress, mental health and GPA. The correlations I computed (step-0 and step-1) show which questions are worth asking. ### 1. How strongly does daily usage track mental health and stress, and is there a threshold where wellbeing drops sharply? Across all 4,500 students, daily usage hours have a strong negative correlation with the Mental_Health_Index (**-0.85**) and a strong positive correlation with Perceived_Stress_Score (**0.754**). The average usage is 5.3 hours a day. These are the strongest relationships in the data. The follow-up is whether the decline is gradual or has a tipping point, for example above 6–8 hours. A tipping point would tell counselors where a recommended limit should sit. The data is observational, so this shows association, not cause. ### 2. Is sleep the pathway from heavy usage to lower GPA? Usage correlates negatively with sleep duration (**-0.716**) and with GPA (**-0.71**). Sleep duration correlates positively with GPA (**0.595**), and sleep quality with mental health (**0.662**). Together these suggest a chain: more usage, less sleep, lower GPA and worse mental health. The question is how much of the usage–GPA link remains once sleep is held constant. Late_Night_Usage (a boolean) is the natural variable to test. If late-night use matters beyond total hours, the advice should be about timing, not just limits on daily hours. ### 3. Which student segments and platforms are most at risk, and does social comparison add risk beyond usage time? The dataset has several segment variables: age, gender, academic level, primary platform, device type, and Social_Comparison_Frequency. It also has an Overall_Impact label. This lets you ask: - Do some groups or platforms have worse outcomes at the same usage level? - Does frequent social comparison predict stress or mental health after controlling for hours? Pre-aggregated breakdowns exist for most of these dimensions. They would show whether outreach should target particular segments or the comparison behavior itself. I haven't run these breakdowns yet, so this is a proposed question, not a finding. ### Caveats - All the correlations are associations. Nothing here shows that usage causes the outcomes. Heavy use could also be a symptom of distress. - The correlations are unusually strong for survey-type data, so check how the measures were collected before relying on them.