Flay 1995.

Methods	Country: USA Site: 340 classes in 6 school districts with 35 Los Angeles and 12 San Diego schools. 'Television, School and Family project' (TVSFP). Focus: Tobacco Design: Cluster RCT (excluded from analysis).
Participants	Baseline: 7352 (6695 (91%) indicated gender, race and smoking status). Age: beginning of 7th grade (approximately 12 yrs). Gender: 49% M Ethnicity: 35% H, 33% W, 14% A‐A, 17% Other. Baseline smoking data: never‐smokers: intervention N = 112, control N = 81.
Interventions	Category: Social influences vs control. Programme deliverer: Trained health educators. Intervention: (a) correction of misperceptions about tobacco usage; (b) awareness of peer influences to smoke; (c) development of peer resistance skills; (d) awareness of family influences to use tobacco; (e) development of media influences resistance skills; (f) social and physiological effects of smoking; (g) development of decision‐making skills. Control: No intervention.
Outcomes	Self reported smoking for the past week (test‐retest stability 0.26 between waves B and C, and 0.31 between waves C and D); ever‐use in lifetime (test‐retest stability 0.71 between waves B and C, and 0.72 between waves C and D). Follow‐up: End of grade 7, 1 yr post‐intervention, 2 yrs post‐intervention.
Notes	Quality of intervention delivery: Numerical results of process analysis not stated; "Instructors completed delivery process questionnaires daily, weekly and immediately post program. Classroom teacher observers were surveyed weekly. The school staff was interviewed during the week immediately following the class session". Parents signed when student‐parent homework was complete; the authors commented "Fidelity of implementation was assured through curriculum delivery by trained health educators [but] Unfortunately the television programming was poorly executed and there was significant variability in the integrity of classroom program delivery". Statistical quality: Was a power computation performed? No. Was an intention‐to‐treat analysis performed? Not stated. Was a correction for clustering made? Yes. Were appropriate statistical methods used? Results were adjusted for clustering using ML3 multilevel analysis programme for unbalanced data that uses iterative generalized least‐squares estimation.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	"Within each of two counties (Los Angeles and San Diego) we assigned entire schools to conditions (22) using a randomised multi‐attribute blocking approach developed by Graham et al". Method of randomisation not stated. Clusters: Schools. Cluster constraint: Multi‐attribute blocking. Baseline comparability: No differences at pretest in smoking rates across conditions.
Allocation concealment (selection bias)	Unclear risk	No statement
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	No statement
Incomplete outcome data (attrition bias) All outcomes	Unclear risk	53% attrition at 2 yrs, with higher attrition among African‐Americans, and students with lower school grades, but there was no differential attrition across groups.
Selective reporting (reporting bias)	Low risk	No selective reporting