Sleepmore in Seattle: Later school start times are associated with more sleep and better performance in high school students

Adolescents show a biologically driven delay in sleep timing during puberty due to changes in circadian period, reduced morning light sensitivity, and altered homeostatic sleep pressure. These changes, combined with early school start times, reduce total sleep below recommended levels (8–10 hours). Delaying secondary school start times has been proposed to increase sleep, but prior evidence largely relied on self-reports; objective evidence that a district-wide delay increases daily sleep was lacking. Because sufficient sleep supports learning and memory, later starts might also improve academic performance, yet causal links in field settings remain unclear. The Seattle School District’s shift from 07:50 to 08:45 a.m. in 2016–2017 provided a natural pre-/post opportunity to objectively test whether delaying start times lengthens sleep and whether this aligns sleep timing with circadian biology, reduces sleepiness, improves grades, and enhances attendance.

Prior work documents adolescent circadian delay and reduced sleep with early schedules (e.g., Carskadon et al.; Wright et al.). The American Academy of Pediatrics recommends later starts. Survey-based studies suggest longer reported sleep with delays, but objective actigraphy evidence for district-wide changes was lacking. Related findings include: shortened sleep and delayed circadian phase when moving to earlier starts (Carskadon et al., 1998); modest sleep gains (~10 min) after a 45-min delay in a Singapore all-girls school (Lo et al., 2018), potentially due to design differences; and experimental middle school interventions showing ~1 hour of increased sleep with a 1-hour delay over a week (Lufi et al., 2011). Reviews (Wahlstrom et al., 2014; Minges & Redeker, 2016) describe health and academic benefits associated with later starts. Evening screen use can delay sleep onset (Chinoy et al., 2018).

Design: Pre-/post natural experiment in Seattle Public Schools comparing spring 2016 (pre; start 07:50) vs spring 2017 (post; start 08:45). Independent samples in each year from the same grade, classes, schools, and season. Setting and participants: 10th-grade Biology students at two Seattle public high schools: Roosevelt High School (RHS) and Franklin High School (FHS). Actiwatches assigned to a subset in each section to represent gender and underrepresented minorities. Informed assent/consent obtained; approvals from University of Washington Human Subjects Division and Seattle Public Schools. Data collection: Students wore Actiwatch Spectrum Plus devices for 2 weeks (15-s epochs). Students pressed event markers at sleep onset/offset and completed daily online sleep diaries. One-time surveys included Epworth Sleepiness Scale, BDI-II, Horne-Östberg and Munich Chronotype Questionnaires. Demographics (sex, race, birthdate, commute, transport) collected; 2017 cohort also reported any new pre-school activities (few cases). Sleep/light/activity processing: Actiware v6 used to define sleep intervals; activity and light exported for analysis in R/Python/Prism. Activity and light data binned into 10-min intervals to compute mean 24-h waveforms for school vs nonschool days. Sleep data cleanup: Three sources (Actiware intervals, diary, event markers) cross-validated. Procedure: (1) visual inspection of actograms to correct obvious miscalls; (2) identify >1 h discrepancies between sources; (3) adjudicate discrepancies via actogram review. Of discrepancies >1 h, 77% due to student error, 19% watch error, 4% undetermined; resulting watch-defined onset/offset error ≤5% after step 1. Inclusion criteria: School nights: Sunday–Thursday; nonschool nights: Friday, Saturday, and night before Memorial Day. Exclude nonschool data if ≥1 nonschool night missing (out of 4); exclude school data if ≥5 school nights missing (out of 10); remove entirely if both criteria unmet. Each student contributed ≥5 school nights and ≥3 nonschool nights when included. Light handling: Used “white light” channel; excluded readings during sleep, off-wrist, or <1 lux due to measurement error/occlusion. Computed times of first and last exposure to ≥50 lux per day (physiologically relevant for melatonin suppression) to avoid distributional and device bias. Academic and attendance data: Second-semester science class grades obtained from teachers (absolute grades). School-wide period 1 absences and tardies provided by district for 2016–2017; predicted 2017 counts based on 2016 rates adjusted for enrollment for χ² tests. Statistical analysis: Activity waveforms: two-way ANOVA (year × time of day) with Sidak’s multiple comparisons; similar approach infeasible for raw light intensities due to non-normality. Times of first/last ≥50 lux: two-way ANOVA (year × day type). Sleep parameters (onset, offset, duration): non-normal; Wilcoxon signed-rank tests with Bonferroni correction (P < 0.017 threshold). Social jet lag: difference between mid-sleep on nonschool days (after oversleep correction) and school days. Effect sizes for Wilcoxon computed as U/(N1×N2). Generalized linear models (binomial family) with year as dependent variable tested differences across school, academic performance, mood (BDI-II), chronotype, sleepiness, and one sleep metric at a time to avoid multicollinearity; model selection by AIC; final model included school, academic performance, mood, chronotype, sleepiness, and weekday sleep offset. Attendance (period 1) differences assessed by χ² tests for FHS and RHS separately.

- Daily sleep duration increased by a median of 34 minutes on school days after the start time delay (Wilcoxon, P = 0.0007; effect size = 0.353). Median sleep rose from 6 h 50 min (2016) to 7 h 24 min (2017). - Sleep offset on school days was 44 minutes later in 2017 (Wilcoxon, P < 0.0001; effect size = 0.194); sleep onset showed a trend but was not significantly different at the multiple-comparison threshold. - No significant differences between years in sleep onset, offset, or duration on nonschool days. - Social jet lag decreased: median 1.60 h (2016, n=81) to 1.25 h (2017, n=76); Wilcoxon, P = 0.0118; effect size = 0.616. - Light exposure timing on school days: first daily exposure to ≥50 lux occurred later in 2017 than 2016 (two-way ANOVA year effect F(1,331)=18.2, P < 0.0001), with no year difference in last daily exposure (F(1,331)=6.2, P = 0.0136 overall year effect but Sidak comparisons showed no between-year difference on school days for the last exposure). On both years, first/last ≥50 lux exposure occurred later on nonschool than school days. - Generalized linear models identified significant between-year differences in: weekday sleep offset (P = 2.8×10^-5; median2016 = 06:24, median2017 = 07:08), academic performance (P = 0.0261; median grade 77.5% in 2016 vs 82% in 2017; ≈4.5% increase), and daytime sleepiness (Epworth median 7.0 in 2016 vs 6.0 in 2017; P = 0.0370). School, sex, depression index (BDI-II), and chronotype were not significant in final/best models. - Attendance and punctuality: At Franklin High School (economically disadvantaged, higher minority representation), first-period absences and tardies were significantly reduced in 2017 vs 2016 (χ², P < 0.0001). At Roosevelt High School, no significant change. - N contributions: school days n = 94 (2016) vs 84 (2017); nonschool days n = 81 (2016) vs 76 (2017). Nap counts were similar (2016: 152 total, 0.6 per student; 2017: 150 total, 0.56 per student).

Delaying high school start time from 07:50 to 08:45 a.m. aligned students’ wake times with their circadian biology, yielding longer sleep on school days without later bedtimes sufficient to offset the gains. The reduced social jet lag indicates better synchronization between school-day sleep schedules and intrinsic circadian timing. Lower daytime sleepiness and higher grades are consistent with improved alertness and potential learning benefits, though causality for academic performance cannot be definitively established in this design. The attendance and punctuality improvements concentrated at the more economically disadvantaged school (FHS) suggest that later starts may help mitigate socioeconomic disparities in educational engagement. Light-exposure analyses show the delay primarily shifted morning light exposure later, with no evidence of prolonged evening bright light, mitigating concerns of circadian delay from evening light. Findings corroborate and extend prior research by providing district-wide, objectively measured actigraphy evidence of increased sleep after a start-time delay. The results support later school starts as a public health and educational policy to improve adolescent sleep, alertness, and potentially academic outcomes.

A district-wide 55-minute delay in high school start times produced objectively measured gains of roughly half an hour of sleep on school days, reduced social jet lag and sleepiness, modestly higher grades, and improved first-period attendance and punctuality in an economically disadvantaged school. These results demonstrate that later starts help realign school schedules with adolescent circadian biology and move students toward recommended sleep amounts. Future work should employ controlled, multi-district and longitudinal designs; examine long-term academic and health outcomes; assess impacts across diverse subgroups; and pair schedule changes with sleep-hygiene interventions (e.g., reducing evening screen exposure) to maximize benefits.

- Pre-/post natural experiment without a concurrent control group limits causal inference; cohort differences between years may contribute. - Independent samples across years (not longitudinal within-subjects); potential unmeasured confounders (e.g., schedule changes, extracurriculars). - Grades were absolute and teacher-assigned; potential implicit biases toward/against schedule change. - Sample limited to two schools and 10th-grade science classes; generalizability may be limited. - Actiwatch light sensors have known inaccuracies at low intensities and off-axis; light exposure analyses used thresholds to mitigate but not eliminate bias. - Attendance analyses limited to first-period school-wide data; improvements not observed at both schools. - Ethnicity not modeled directly, though schools differed demographically; sex differences not detected but power may be limited. - Some pre-school activities added after delay (2017) were not modeled separately (reported to be few). - Sleep measures based on 2-week recordings; nonschool nights fewer in count; weekend oversleep corrected for but residual variability remains.