Hoehn 2015.

Methods	Non‐randomized cross‐over trial
Participants	Eligibility: 8th graders from 32 middle schools in a major metropolitan school district; any adolescent who completed the Youth Risk Behavior Survey was eligible Participants: 32,980 students (50.2% female, 58.2% non‐white)
Interventions	Earliest start time (07:20‐07:30), early start time (07:40‐07:55), late start time (08:00‐08:10)
Outcomes	Sleep duration Outcome measurement: Statewide questionnaire (modified Youth Risk Behavior Survey), administered once annually over three years Respondents categorized by school start time, analyses by multilevel multinomial logistic regression
Notes	Funding sources: supported by Scholarship for Studying Abroad from China Scholarship Council, Distinguished Young Academics Fund from East China Normal University
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	High risk	Non‐randomized trial; high risk by default
Allocation concealment (selection bias)	High risk	Non‐randomized trial; high risk by default
Blinding of participants and personnel (performance bias) All outcomes	High risk	No blinding of participants or personnel mentioned; participants knew their start time
Incomplete outcome data (attrition bias) All outcomes	Unclear risk	No discussion of attrition
Selective reporting (reporting bias)	Low risk	Only captured one outcome; reported that one outcome
Other bias	Low risk	No evidence of other biases
Similarity of baseline measures (RCT/CBA only)	High risk	No baseline measure
Similarity of baseline characteristics (RCT/CBA only)	Low risk	Controlled for gender, race, SES
Diffusion of treatment effects (RCT/CBA only)	Low risk	Because of the nature of school start times, diffusion is not a possibility between groups