Student Performance Dashboard

Sleep, Screens & SuccessStudent performance dashboard

Methodology & data notes

500 students, 25 fields covering sleep, screen use, lifestyle, study and attendance, and cumulative GPA (CGPA, 4.0 scale): student_sleep_screen_time_academic_performance.csv.
Students are identified only by pseudonymous IDs. The data is a single cross-sectional snapshot, with no time dimension.

Gender: 105 inconsistent labels normalised (M / male → Male, F / female → Female).
Bedtime: times after midnight shifted +24 h so they sort after 23:00. Correct median 23:42; a naive average would read 12:47.
Missing values are kept and flagged, never imputed. GPA metrics exclude the 7 students without CGPA. Sleep quality (6), energy (15), concentration (14) and device-free routine (9) are excluded only from the metrics that use them. Every tile shows its own n.

27 students (5.4%): weekday sleep < 4 h or > 10 h, daily screen > 14 h, study > 60 h/week, or attendance < 50%. Use Exclude outliers to drop them. Results barely change (sleep β 0.41 → 0.40).

Levels: Low < 25 ≤ Moderate < 40 ≤ Elevated < 55 ≤ High. If sleep quality is missing, the other weights are renormalised to 100.
v2 replaced caffeine with low study hours, because caffeine showed no effect once sleep was controlled. Correlation with CGPA strengthened from r = −0.47 to −0.56.

Drivers chart: Pearson r with CGPA on all 493 students with CGPA, plus standardised β from multiple regression. It does not change with filters.
Sleep-adjusted chronotype: group means of CGPA evaluated at the average sleep of shown students, using a pooled within-group slope (ANCOVA).
What-if slider: simple linear slopes of sleep, sleep quality and CGPA on pre-sleep screen minutes (all students), applied to each shown student's own reduction. The ± range is the 95% interval of the slope.
Dot charts show means with 95% confidence intervals (1.96 × SE).

Deltas on KPI tiles compare the current view with all 500 students. Targets are proposals pending owner sign-off.
Groups under 30 students are faded and badged “low n”. Heatmap cells under 10 are hidden (hatched).
Profile actions are rule-based suggestions with thresholds pending owner sign-off. They are not diagnoses.

Every KPI was recomputed independently in Python (validate_dashboard.py) for no filters and 3 random filter combinations. All 368 value, n and displayed-text checks matched. Details: validation_report.md.

Associations, not causes. The data is observational and self-reported. Sleep, study and attendance together explain about 28–34% of CGPA variance; most variation is unexplained.
Seven majors and several subgroups are small. Treat differences within small groups as hypotheses.