Buka 1999.

Methods	Non‐randomised controlled trial, unit of allocation = community.
Participants	Rhode Island; USA. Three communities; one experimental and two control sites. Experimental = containing 51 bars Control site A = containing 26 bars Control site B = containing 26 bars
Interventions	Server training CAAIPP alcohol server training. Twenty‐four training courses were held, each lasting five hours, with 5‐15 servers attending each course. The training curriculum was developed from the National Highway Traffic Safety Administration. The training emphasised training by peer servers; each session was co‐facilitated by a server and an alcohol treatment/prevention professional from the community. The purpose of the training was to provide servers with the knowledge and skills required to prevent patrons from becoming intoxicated, prevent service of alcohol to minors, identify and stop service to intoxicated patrons and help prevent injuries to those individuals as well as informing servers of their legal liability if they fail to obey dram shop laws. Control communities were not exposed to CAAIPP training.
Outcomes	Self reported server behaviour, measured by a Desired Server Behaviour Index.
Notes
Risk of bias
Bias	Authors' judgement	Support for judgement
Adequate sequence generation?	Unclear risk	Three communities, one of which was reported as being selected at random to be the experimental site and the remaining two sites were used as controls.
Allocation concealment?	Unclear risk	No information reported.
Blinding? All outcomes	Unclear risk	Not reported.
Confounders	Unclear risk	The communities were reported as being similar in regard to vital statistics, hospital discharge data and police statistics on motor vehicle crashes and arrests. The communities were of comparable size, sociodemographic characteristics, prevalence of alcohol‐related problems and levels of institutional development and community organisation.
Data collection methods	Unclear risk	Self‐reported server behaviour was measured by questionnaire. From a sample of 25 premises from each control site and 50 experimental premises, three servers were randomly selected to complete the questionnaire. Response rates for questionnaires was 68% in the experimental area (31% of these were from trained servers), 72% in control A and 63% in control B.
Withdrawals & dropouts	Unclear risk	N/A
Intervention integrity	Unclear risk	Of 531 servers in the experimental community, 324 (61%) completed the intervention.
Duration of follow‐up	Unclear risk	Data were collected four years after intervention.