Wallin 2002.

Methods	Non‐randomised controlled trial, unit of allocation = area.
Participants	Licensed premises in Stockholm, Sweden. Experimental area = northern part of central Stockholm (˜550 licensed premises). Control area = southern part of central Stockholm (˜270 licensed premises).
Interventions	Server training 'STAD project', a multi‐component community alcohol prevention project initiated in 1996. Main categories: 1) Community mobilisation; 2) Two day responsible beverage service training course; 3) Enforcement of existing alcohol regulations. The server training course targeted restaurant owners, bartenders, servers and doormen. It covered the medical effects of alcohol consumption, information about alcohol laws, server intervention training, other drugs and group discussions.
Outcomes	Observed server behaviour (using pseudo‐drunks). Observed server behaviour ‐ customers who appeared to be under 18 attempted to buy a drink.
Notes	Injury data (police reported violence) were also collected, however not used in this review as it was not possible to attribute changes in this outcome to the server training component of the intervention.
Risk of bias
Bias	Authors' judgement	Support for judgement
Adequate sequence generation?	High risk	Experimental area comprised the northern part of central Stockholm (containing ˜550 licensed premises at baseline) and control comprised the southern part of central Stockholm (containing ˜270 licensed premises).
Allocation concealment?	Unclear risk	Inadequate.
Blinding? All outcomes	Unclear risk	Pseudo‐drunks and "under‐18" patrons were blind to allocation status.
Confounders	Unclear risk	Any differences in baseline characteristics between the two areas are not reported in pseudo‐drunk study. "Under‐18" study reports several differences; thus 23% of intervention premises were nightclubs but only 2% of control premises.
Data collection methods	Low risk	Observations of server behaviour to pseudo‐drunks and "under‐18s" used to assess server behaviour. Violence data were collected from police records for 48 months before and 33 months after.
Withdrawals & dropouts	High risk	The experimental and control areas both remained in the study. (In both intervention and control areas some premises closed and new premises opened; actual figures not supplied.)
Intervention integrity	High risk	Only 37 of the 61experimental bars sampled had trained their staff when the pseudo‐drunk follow‐up was done. An unspecified number had not trained their staff when the "under‐18" follow‐up was done.
Duration of follow‐up	Low risk	Observations conducted at the start of the intervention (1996) and at follow‐ups: pseudo‐drunks three years after intervention (1999); "under‐18s" two years (1998) and five years (2001).