C14 Later school start
narrowed 3 attacks: 2 survived, 1 narrowed · serves Q1 · last change 12 November 2026
Starting the secondary-school day later has let teenagers sleep longer and improved their school results in several studies, but not consistently: when one New York high school started 45 minutes later, students at first reported about 20 minutes more sleep, the gain was gone a year later, grades did not change in the two years after the change, and state exam scores showed no consistent change in the first year after (the only year compared) (Thacher & Onyper 2016).
In practice
If teenagers must be in school very early, a later start is one of the few changes that has added real sleep, and in several studies results rose too. It does not always work: in one school the extra sleep faded within a year and grades did not change. Expect the gains in results to be smaller and less certain than the gains in sleep, and check whether students actually sleep more.
Go deeper
Where it holds
Main limits
- About school start times for teenagers, not about evening screens (candidate K04).
- Seattle data are a before/after comparison of two cohorts, with sleep measured by wrist monitors (Dunster et al. 2018).
- Wake County evidence comes from natural variation in start times, not random assignment (Edwards 2012).
- Grades rose in the second year, not the first, after a 50–65-minute delay in two Minnesota high schools compared with schools that stayed early (James et al. 2023).
- Not consistent: in one New York high school a 45-minute later start gave about 20 minutes more self-reported sleep at first, none a year later (though the share sleeping ≥8 h went 40% → 60% → 50%), absences rose about 5 days/year in the within-student analysis, course failures spiked in the transition year (6.04%) then returned, no change in grades over the two years after, and no consistent change in state exams in the first year after (the only year compared) (Thacher & Onyper 2016; one school, no comparison school).
All limits (11)
Inside the circle: the core of the claim. Outside: every bound and carve-out in the claim file, numbered, with the attack that forced or added it where the record shows one.
forced by a narrowing attackin the original wordingwritten at creation, or added without a narrowing
Warm cards in the list were added by later attacks; plain cards were in the original draft or written at creation.
Secondary-school (roughly ages 11-18) start times, outcomes measured sleep and grades or test scores; not evening screen use (candidate K04) or sleep hygiene programmes.
Written when the claim was created (16 October 2026); not forced by an attack
Evidence on file: Dunster et al. 2018 (Science Advances; Seattle start delayed nearly an hour; median sleep +34 min, median grades +4.5%, attendance improved; pre/post, not randomised); Edwards 2012 (Economics of Education Review; Wake County, North Carolina middle schools; later starts associated with higher test scores, larger for lower-scoring students).
Written when the claim was created (16 October 2026); not forced by an attack
Related lab evidence: Cousins et al. 2019 (sleep restriction impaired adolescents' long-term retention).
Written when the claim was created (16 October 2026); not forced by an attack
Consistency bound (added 2026-10-17, nightly attack; enriched 2026-11-06): Thacher & Onyper 2016 (Sleep 39(2), 271–281; full text read): one public high school in Glens Falls, New York, moved first period from 07:45 to 08:30; self-reported sleep rose by about 20 minutes two months after the change and was back to baseline a year later; the share reporting ≥8 h sleep went 40% → 60% → 50% (a net ~10 pp lasting gain in the upper tail while mean TST rebounded); tardiness fell (unexcused tardy ratio 0.066 → 0.05) and disciplinary incidents fell and stayed lower; absences rose in the within-student analysis (all-absence ratio 0.085 → 0.113, about +5 days/year; same-student subset 12.8 → 17.7 days/year) — a cross-sectional roster count without repeated measures wrongly suggested a tiny decrease; course-failure rate among students who experienced the change spiked to 6.04% in the transition year then returned (4.72%, 4.49%, 6.04%, 4.39%; χ²(3) = 14.31, P = .003); no change in grades in cohort-sequential analyses of 2010–2014 records (two years before and two after), and no consistent change in Regents exam scores when 2011–2012 was compared with 2012–2013 (the only year after the change that was compared).
Added with attack #5 (corrected; it did not narrow the claim) · Thacher & Onyper 2016 · 6 November 2026
One school, no comparison school, self-reported sleep.
Forced by attack #2 · Thacher & Onyper 2016 · 17 October 2026
Kill-condition test survived 2026-10-16 (James et al. 2023, grades prong).
Forced by attack #2 · Thacher & Onyper 2016 · 17 October 2026
Kill-condition test survived 2026-10-22 (Dou, Gihleb, Giuntella & Lonsky 2026, NBER Working Paper 35184, DOI 10.3386/w35184; working paper, not peer reviewed); figures corrected 2026-11-12 to the October 2026 revision of the paper: after California's statewide mandate (SB 328; middle schools no earlier than 8:00, high schools no earlier than 8:30), the share of high-school students reporting at least 8 hours' sleep rose 3.6 percentage points (13%) against other states, and teenagers' (ages 15–18) weekday diary sleep about 28 minutes (0.46 hours).
Added with attack #4 (survived; it did not narrow the claim) · Dou, Gihleb, Giuntella & Lonsky 2026 (preprint) · 22 October 2026
Wording changed later: Added with attack #6 (corrected; it did not narrow the claim) · Dou, Gihleb, Giuntella & Lonsky 2026 (preprint) · 12 November 2026.
Test scores now come only from a within-California comparison: schools in counties where the mandate forced a bigger shift (earlier pre-policy starts, proxied by teachers' commute arrival times) against schools in counties that already started late or were rural; the share of students meeting state standards rose about 1.7–1.8 percentage points in maths, English and science in high school and about 1.7 in grade-8 maths and English (0.7 in science, p < .10), growing over 2023–2025.
Added with attack #6 (corrected; it did not narrow the claim) · Dou, Gihleb, Giuntella & Lonsky 2026 (preprint) · 12 November 2026
The revision drops the earlier cross-state comparison of grade-8 scores (about 0.10 and 0.08 standard deviations against matched districts in other states in the May 2026 version), because, in the authors' words, cross-state comparisons of test scores after 2022 cannot isolate the policy; the within-state design does not rule out counties recovering from the pandemic at different speeds.
Added with attack #6 (corrected; it did not narrow the claim) · Dou, Gihleb, Giuntella & Lonsky 2026 (preprint) · 12 November 2026
Sleep self-reported, scores and sleep from different data sets; the kill still fails on the revised paper (more sleep, higher test scores), but on weaker identification.
Added with attack #4 (survived; it did not narrow the claim) · Dou, Gihleb, Giuntella & Lonsky 2026 (preprint) · 22 October 2026
Wording changed later: Added with attack #6 (corrected; it did not narrow the claim) · Dou, Gihleb, Giuntella & Lonsky 2026 (preprint) · 12 November 2026.
Review context: Marx et al. 2017 (Cochrane Database of Systematic Reviews, CD009467; abstract and plain-language summary only): 11 studies, evidence rated very low quality; six studies found more sleep with later starts; four studies of academic outcomes gave mixed results.
Forced by attack #2 · Thacher & Onyper 2016 · 17 October 2026
3 of 11 bounds and carve-outs were forced by a specific attack. 2 attacks hit the boundary and did not move it: #1, #4.
How it changed
The ledger records 3 attacks (and 3 corrections to our notes), but no earlier wording of C14 is on record, so the text cannot be replayed step by step. Only the current wording is shown.
Starting the secondary-school day later has let teenagers sleep longer and improved their school results in several studies, but not consistently: when one New York high school started 45 minutes later, students at first reported about 20 minutes more sleep, the gain was gone a year later, grades did not change in the two years after the change, and state exam scores showed no consistent change in the first year after (the only year compared) (Thacher & Onyper 2016).
What would change this
What would kill it: A well-controlled study of a later secondary-school start finds no gain in measured sleep, or more sleep but no gain in grades or test scores. Added as a claim on 16 October 2026, promoted from a candidate belief in the evidence library.
Attack record
3 attacks: 2 survived, 1 narrowed. Counts of attacks, not a score.
Full attack list (3)
Two US high schools moved their start 50–65 minutes later while three nearby schools stayed early; following about 2,150 students for two years tests whether the later start raised grades. If grades did not move, C14's kill-condition (more sleep but no gain in grades) would be met. It did: grades rose in the second year.
Strongest reading and verdict reasoning
Steelmans to: C14 dies if a well-controlled study of a later secondary-school start finds more sleep but no gain in grades; this cohort is the one in which earlier START work showed longer sleep after the delay (Widome et al. 2020, as cited by the authors), with comparison schools that kept early starts.
Why: Broadening attack run 2026-10-16 (broad, cycle 1) Europe/Stockholm. Full text read (PMC10257742; not stored). Two schools delayed start by 50–65 minutes in fall 2016; three comparison schools kept early starts; students in grade 9 in 2015–16; outcomes from school records. GPA difference-in-differences: no difference in the first follow-up year (0.01 and −0.0005, ns); second year +0.17 (95% CI 0.11–0.23) and +0.14 GPA points (weighted measure; 0.08 and 0.07, p < .05, unweighted); also fewer late arrivals and absences and fewer behaviour referrals. The sleep gain is reported from the same cohort's earlier paper, not re-measured here. The authors report larger GPA gains for students without free or reduced-price lunch and with a college-graduate parent (e.g. 0.21 vs −0.02 weighted GPA at follow-up 2a), which runs against Edwards 2012's larger effect for lower scorers; logged as a lead, not a narrowing (subgroup, not the claim's wording). Not randomised; two treated schools. Not a preprint.
When one New York high school moved its start 45 minutes later, students reported only about 20 minutes more sleep at first and none a year later, and grades and state Regents exam scores did not improve in the two years after the change, so a later start does not reliably bring longer sleep and better results.
Strongest reading and verdict reasoning
Steelmans to: C14 asserts both links (later start → more sleep → better results) without qualification; this is a longitudinal study of a whole school with school records for two years on each side of the change, and it fails on both links; a Cochrane review (Marx et al. 2017) rates the evidence on academic outcomes as very low quality, with mixed results across four studies.
Why: Nightly-attacks 2026-10-17 Europe/Stockholm. Full text read (PMC4712391; stored privately). Glens Falls, New York, a walking district; first period moved from 07:45 to 08:30 from school year 2012–13; surveys at baseline (May 2012) and in November 2012 and May 2013 (410 and 372 students in grades 9–12 at the two follow-ups); school records 2010–2014. Self-reported (PSQI) total sleep: about 20 minutes longer at the first follow-up, back to baseline by May 2013 as bedtimes moved later; tardiness and disciplinary incidents fell and stayed lower. Grades: cohort-sequential mixed models found no change tied to the start time ("no evidence to suggest that a change in school start times from earlier to later was associated with either improvement or decline in academic performance"); cross-sectional comparisons found no differences in grades 9, 10 and 12 and a small increase in grade 11; on Regents exams 2 of 20 subject tests differed, both higher before the change. Kill withheld: one school, no comparison school, self-reported sleep, cohort comparisons — C14's kill-condition asks for a well-controlled study, and James et al. 2023 (with comparison schools) and Dunster et al. 2018 (sleep measured by wrist monitors) found gains. Narrowed: the gains are shown in some settings, not consistently. Context (abstract and plain-language summary only): Marx et al. 2017, Cochrane review of 11 studies, evidence very low quality; four studies of academic outcomes with mixed results. Not a preprint. Status alive → weakened. Claim and Bounds rewritten same turn; brief and the practical rule rewritten.
- #4 survived 22 October 2026 · Dou, Gihleb, Giuntella & Lonsky 2026 (preprint) · target: kill-condition
California's SB 328, the first statewide mandate for later secondary-school starts (middle schools no earlier than 8:00, high schools no earlier than 8:30, binding in about 80% of districts), is the largest well-controlled test of a later start on record; if teenagers slept more but test scores did not rise against comparable districts in other states, C14's kill-condition (more sleep but no gain in grades or test scores) would be met. It was not: sleep rose and grade-8 maths and English scores rose.
Strongest reading and verdict reasoning
Steelmans to: C14 dies if a well-controlled study of a later secondary-school start finds more sleep but no gain in grades or test scores; a statewide mandate evaluated by difference-in-differences against matched districts in other states, with event-study pre-trends, a rural-exemption triple difference and a within-state design, is the strongest such test available, and earlier doubts (no consistent change in a single school, Thacher & Onyper 2016; mixed review findings) make a null plausible.
Why: Nightly-attacks 2026-10-22 Europe/Stockholm. DOI checked on Crossref (title, four authors, NBER, May 2026). Full text read: NBER May 2026 version as page images (its text layer does not extract; not stored) and the May 2025 IZA version as text (EconStor, stored privately as ). May 2026 version: YRBS share of high-school students sleeping at least 8 hours up 13% (pp. 3–4); ATUS weekday time diaries, ages 15–18, sleep +0.712 hours (SE 0.241) and share sleeping at least 8 hours +0.122 (SE 0.058) (Table 3, p. 27); SEDA grade 8, 2015–2024 without 2020–21: two-way fixed effects maths +0.045 SD (SE 0.012), English +0.033 (SE 0.011); matched DiD maths +0.096 (SE 0.015), English +0.077 (SE 0.013), all p < .01 (Table 5, p. 33); pre-period event-study coefficients small and not significant (p. 32); urban–rural triple difference maths +0.09, English +0.14 (p. 33). Kill fails: more sleep came with higher test scores. Caveats: working paper, estimates fell between versions (May 2025 matched DiD on grades 3–8 was 0.169 and 0.149 SD); sleep self-reported, measured mostly in high-school students while scores are grade 8; post-period follows COVID-19 disruptions; no grades outcome. Not a duplicate (prior entries: James et al. 2023 survived on the grades prong; Thacher & Onyper 2016 weakened and corrected). Claim field unchanged; Bounds, brief C14 paragraph and the practical rule Limits record it; recheck item added to in the attack method, (the preprint rule). Status unchanged (weakened).
Corrections to our notes
These fix a number or a fact in our notes. They are not attacks and do not count toward the floor of 3 survivals.
- — The claim overstated Thacher & Onyper 2016: grades were tracked for two years after the start-time change, but state Regents exam scores were compared for only the first year after (2011–2012 vs 2012–2013).
Details
What was checked: fact in claim text (exam-score follow-up window), not a new attack on the claim's force.
Why: Correction 2026-10-17 Europe/Stockholm (not an attack run). Full text re-read (/ PMC4712391). Methods: school records for tardiness, absences, discipline and academic performance covered 2010–2014 — "2 years prior and 2 years following the change"; grades used a cohort-sequential comparison across those years plus a cross-sectional comparison. Standardized assessment scores: "Scores on standardized assessments (Regents exams) in 2011–2012 (before schedule change) were compared with scores in 2012–2013 cross-sectionally" — only the first year after. Of 20 subject tests, 2 differed and both were higher before the change. Scope, direction and kill-condition unchanged; Status stays weakened. Not a preprint.
- — Checked against the full text of the study: facts behind the consistency bound (absences, transition-year failures, ≥8 h sleep share).
Details
What was checked: facts behind the consistency bound (absences, transition-year failures, ≥8 h sleep share); no kill-condition test.
Why: Nightly-attacks 2026-11-06 Europe/Stockholm (the targeting rules queue). Full text: all-absence ratio 0.085 → 0.113, F(1,505) = 35.47, P < .001 (~+5 days/year); same-student subset n = 487, 12.8 → 17.7 days/year; course-failure rates among students who experienced the change 4.72%, 4.49%, 6.04% (transition year), 4.39%, χ²(3) = 14.31, P = .003; ≥8 h sleepers 40% → 60% → 50%; unexcused tardy ratio 0.066 → 0.05. Not a duplicate (Bounds previously omitted absences, the failure spike and the ≥8 h nuance); not a restatement of the existing sleep-fade / null-grades / mixed-Regents sentences. Kill withheld: at one year mean self-reported sleep was already back at baseline and academics did not rise — already the consistency bound; absences moving the wrong way strengthen the "not consistent" reading but do not by themselves kill while Dou/Dunster-type gains remain on file. Claim text, Status and kill line unchanged. Not a survival; does not count toward the floor; does not reset C14's fruitless streak. C14 returns from rest under the targeting rules(a) on this verified full-text 3g flag.
- — The October 2026 revision of Dou et al. drops the cross-state comparison of grade-8 test scores that C14's limits quoted (about 0.10 and 0.08 SD against matched districts in other states), because the authors now say cross-state comparisons after 2022 cannot isolate the policy, and reports test scores only within California (share of students meeting state standards up about 1.7–1.8 percentage points where the shift…
Details
What was checked: facts behind the 2026-10-22 kill-condition survival (which estimates the working paper now reports); no new kill-condition test.
Why: Nightly-attacks 2026-11-12 Europe/Stockholm, recheck of a working paper under the preprint rule (a closer reading of the study's version caution is superseded by this entry). Revised full text read (89 pp.; title page "May 2026, Revised October 2026"). YRBS (Table 2, PDF p. 22): share sleeping ≥ 8 h +0.036 (SE 0.017; 13% of the mean 0.271); ≥ 7 h +0.070 (SE 0.017). ATUS ages 15–18, weekdays, metro areas (Table 3, PDF p. 24): sleep hours +0.464 (SE 0.207; "about 28 minutes"), share ≥ 8 h +0.097 (SE 0.044). Academic outcomes (PDF p. 5; Table 5, PDF p. 36): within-California design, school fixed effects, county exposure proxied by teachers' commute arrival times in the ACS (bottom three quartiles vs top quartile plus rural counties for high schools), CDE records 2014–15 to 2024–25 without 2019–20 and 2020–21, SEs clustered by county; share meeting or exceeding standards, high school maths +1.823 (SE 0.686), English +1.745 (SE 0.591), science +1.729 (SE 0.627); grade 8 maths +1.700 (SE 0.483), English +1.697 (SE 0.539), science +0.693 (SE 0.395, p < .10); event studies flat before 2022 and rising through 2025 (PDF pp. 30–34). The SEDA cross-state grade-8 analysis, the matched DiD and the urban–rural test-score triple difference of the May 2026 version are gone; the authors write that "the uneven recovery from the pandemic across states means that cross-state comparisons of test scores after 2022 cannot isolate the policy" and that the within-state design "does not rule out differences in recovery across counties that happen to be correlated with pre-policy start times" (PDF pp. 33–35). The kill-condition (more sleep but no gain in grades or test scores) still fails on the revised paper, so the 2026-10-22 survival stands as logged; the 2026-10-22 entry is not edited. Scope, direction, kill-condition and Status (weakened) unchanged; claim text unchanged. Not a survival; does not count toward the floor; fruitless streak unchanged.
Evidence
9 linked sources: 7 supports, 2 limits (first-pass draft, pending human review). supports: Dunster et al. 2018, Edwards 2012, James et al. 2023, Dou et al. 2026, Albrecht et al. 2026, Bowers & Moyer 2017, Alfonsi et al. 2023 · limits: Thacher & Onyper 2016, Marx et al. 2017