2. Results ‐ server training (behaviour).

Buka 1999	SELF‐REPORTED SERVER BEHAVIOUR Alcohol serving practices ‐ this was measured in each community using a Desired Server Behaviour Index (DSBI) (score ranged from 1 to 5). The higher the score the more desirable the behaviour. Mean DSBI (+/‐ SD) for overall server behaviour; Experimental community = 3.59 (+/‐ 0.74) Control community A = 3.59 (+/‐ 0.61) Control community B = 3.24 (+/‐ 0.65) Significance test; F=2.96, P=0.06.
Gliksman 1993	OBSERVED SERVER BEHAVIOUR TO PSEUDO DRUNKS Measured using a behaviour score based on observations of six scenarios (the higher the score the more desirable the behaviour). (Exact figures are not reported in the report text, but the following estimates were read from a graph). Experimental sites behaviour score increased from ˜15 to 21.5 pre to post intervention. Control sites behaviour score changed from ˜16.5 to 16.4 pre to post intervention. Significance test; F=8.73, P<0.01.
Graham 2004	OBSERVED AGGRESSION EXHIBITIED BY PATRONS (average number of incidents per observation) 1) Consistent rating of severe physical aggression by all raters, definite intent Experimental bars; decreased from 0.053 to 0.035 Control bars; increased from 0.007 to 0.060. Significance test; t= 5.23, df=28, P<0.001. 2) All severe aggression plus consistent rating of moderate physical (with or without verbal aggression), definite intent (average number of incidents per observation) Experimental bars; decreased from 0.134 to 0.101 Control bars; increased from 0.075 to 0.126. Significance test; t= 1.87, df=28, P=0.071. OBSERVED AGGRESSION EXHIBITIED BY STAFF (average number of incidents per observation) 1) Consistent rating of severe physical aggression by all raters, definite intent 'Frequencies too low for analyses. 2) All severe aggression plus consistent rating of moderate physical (with or without verbal aggression), definite intent Experimental bars; increased from 0.029 to 0.056. Control bars; increased from 0.014 to 0.053. Significance test; t= 1.19, df=28, P=0.243.
Holder 1994	Not measured.
Howard‐Pitney 1991	OBSERVED SERVER BEHAVIOUR Mean number of interventions made by servers were calculated for eight different responsible interventions, the overall mean for all eight interventions (the higher the mean value the more desirable the server behaviour); Experimental bars = 0.95 Control bars =1.26 Confidence intervals and results of significance test are not presented however, the authors report that 'no differences were observed between treatment and control servers on any intervention or on a sum average of eight possible interventions'.
Johnsson 2003	BEHAVIOUR OF PATRONS Change in BAC(mg%) (and 95% CIs) between baseline and follow‐up; Experimental bars= ‐0.004% (‐0.012 to 0.004) Control bars = +0.007% (‐0.001 to 0.015). Mean difference in BAC between experimental and control bars = ‐0.011% (95% CI 0.022 to 0.000). In the experimental group 40% of tested patrons had a BAC greater than 0.1% before the training and 39% after. In the control group the corresponding figures were 34% before and 41% after. The difference between these changes was not significant (P = 0.12, one‐tailed, 95% CI ‐0.45 to 1.10).
Krass 1994	BEHAVIOUR OF PATRONS 1) Mean BAC (mg%) of patrons This increased from 0.055 (95% CI 0.049 to 0.065) to 0.069 (95% CI 0.058 to 0.078) over the study period in the experimental sites, and increased from 0.057 (95% CI 0.050 to 0.078) to 0.058 (95% CI 0.050 to 0.066). 2) Total consumption of alcohol (gm) On experimental premises this increased from 62.4 (95% CI 50.5 to 74.4) to 69.3 (95% CI 56.9 to 81.6), and decreased from 79.0 (95% CI 82.9 to 95.1) to 67.9 (95% CI 56.7 to 79.1) in control premises. The authors report that 'no significant differences were found in mean BAC and total consumption of alcohol between experimental and control sites at pre and post level'. 3) Proportion of patrons with a BAC over 0.10mg% This increased from ˜0.17% to ˜0.27% in intervention sites and reduced from ˜0.23% to 0.2% in the control sites (exact figures are not presented in the report text, but the following estimates were read from a graph). No confidence intervals or significance test results presented for this outcome.
Lang 1998	BEHAVIOUR OF PATRONS 1) Drink driving offences No quantitative data presented. The authors report that 'the downward trend in drink driving offences from intervention premises leading up to the project was continued during the evaluation period, while the figure for the control sites remained relatively unchanged. However, the number of drink driving cases from both intervention and control premises were too few to permit any meaningful evaluation'. 2) Percentage of tested patrons with a BAL(mg%)> 0.15 This reduced over the study period, with the decline greater for experimental sites (17.4% to 5.3%) than control (10.1 to 3.7%), this is not significant (P=0.389). 3) Percentage of tested patrons with a BAL>0.08 This decreased from 52% to 26.9% in experimental sites and decreased from 34.8% to 24%, this rate of decline is significantly greater (P<0.029) for the experimental than for the control group. SELF‐REPORTED SERVER BEHAVIOUR Changes in the average ratings in mean score of the adoption of responsible service policies over the pre and post periods were reported as not statistically significant (full results of significance test is not presented). Total score increased from ‐0.7 to 0.9 in intervention sites and remained unchanged at ‐1.8 in control sites (maximum possible score=+2, minimum possible score= ‐2). OBSERVED SERVER BEHAVIOUR Reported that there was no difference between experimental and control in terms of refusal of service to intoxicated pseudos. In the experimental group, 1 out of 11 visits and 3 out of 14 visits were refused service in the pre and post period respectively. In the control group, 1 out of 14 visits were refused service in both the pre and post period. Authors report that 'no further analyses were undertaken'.
McKnight 1991	SELF‐REPORTED SERVER BEHAVIOUR (trained servers only) 1) Serving practices, mean score (+/‐sd); Pre = 3.13 (+/‐ 0.67) Post = 3.50 (+/‐ 0.68) Significance test; diff=0.57, t=11.90, P<0.01 2) Serving policies mean score (+/‐sd); Pre =0.58 (+/‐ 0.12) Post = 0.65 (+/‐ 0.11) Significance test; diff=0.61, t=6.65, P<0.01 OBSERVED SERVER BEHAVIOUR 1) Percentage change between pre and post server 'intervention level': a) Intervention level = 'None' (servers make no attempt to intervene); Experimental = ‐12.5% Control = ‐0.8% b) Intervention level = 'Partial' (servers provide drink requested but make some attempt at intervention); Experimental = +10.5% Control = +1.7% c) Intervention level = 'Full' (servers refuse to serve any alcoholic beverage); Experimental = +1.9% Control = ‐0.7% 2) Mean score of server intervention (the higher the score the more desirable); Mean score in experimental sites increased from 0.19 to 0.34 (diff = 0.15, F=10.42, P<0.01) between the pre and post periods. Mean score in control sites remained at 0.22 (diff = 0.00, F=0.01, P=0.97) between the pre and post periods. Significance test of the difference between the intervention effects in the experimental and control sites; F=6.70, df=1/207, P=0.01). OBSERVED 'REAL' PATRON INTOXICATION Amongst the experimental sites the mean intervention level increased from 0.03 before, to 0.22 after (F=4.27, df=1/127, P=0.04), for the comparison sites remained unchanged at 0.07 (F= 0.87, df=1/167, P=0.35) across the periods.
Russ 1987	OBSERVED SERVER BEHAVIOUR Trained servers on average were reported as attempting a greater frequency of intervention than servers without training (P<0.05). BEHAVIOUR OF PATRONS The average exit BAC(%mg) for pseudo patrons served by servers who remained untrained was 0.103 (+/‐ 0.033), while those served by trained personnel had an average BAC of 0.059 (+/‐0.019). The mean difference in exit BACs between pseudopatrons served by trained versus untrained servers = 0.044 (95% CI 0.022 to 0.066). Authors report that the 'BAC levels of pseudopatrons served by trained staff were significantly lower (P<0.01) than those obtained among pseudopatrons prior to training or served by untrained servers in the post period'.
Saltz 1987	BEHAVIOUR OF PATRONS 1) Per capita consumption (number of drinks) Reduced from ˜5.6 to ˜5 in experimental site and ˜6 to ˜5.5 in the comparison site. 2) Rate of consumption (drinks per hour) Reduced from ˜3.5 to ˜2.3 in the experimental site and ˜3.25 to ˜3.75 in the comparison (exact figures are not reported in the report text, but the following estimates were read from a graph). Confidence intervals or results from significance test were not reported. Authors report that 'multivariate linear and logistic regression analyses 'reveal that although absolute consumption and rate of consumption were unaffected by the program, the likelihood of a customers being intoxicated was cut in half'.
Saltz 1997	SELF‐REPORTED SERVER BEHAVIOUR Self‐reported server policy of refusing service to intoxicated patrons (Mean % yes) N. Californian communities Experimental; pre = 3%, post = 19% Control; pre =8%, post = 10% S. Californian communities Experimental; pre 6%, post = 15% Control; pre =6%, post =7% S. Carolina communities Experimental; pre= 7%, post= 8% Control = pre 19%, post 17%. Authors report that 'no statistical difference was found', no further information presented. OBSERVED SERVER BEHAVIOUR Pseudo‐patron survey; responsible service assessed using an intervention score (ranging of low [=bad] of ‐2 to +2 [=good]). N. Californian communities Experimental; pre = 0.17, post = 0.21 Control = pre ‐0.15, post = ‐0.19 S. Californian communities Experimental; pre= ‐0.18, post = ‐0.17 Control = pre 0.15, post 0.16 S. Carolina communities Experimental; pre= 0.17, post= 0.07 Control; pre= ‐0.23, post= ‐0.09. Authors report that 'no statistical difference was found', no further information presented.
Toomey 2001	OBSERVED SERVER BEHAVIOUR Pseudo‐intoxicated purchase attempts Pre‐intervention, the purchase rates were 68.4% and 70.1%, in the experimental and control sites respectively. Post‐intervention, the purchase rate reduced in the intervention site to 40.0% and increased to 72.9% in the control. The relative decline was reported as not statistically significant (t=‐1.17, P=0.27). Refusal of service to pseudo‐intoxicated patrons changed from 83.1% to 80.3% in experimental and from 63.0% to 54.8% in control (t=0.24, P=0.81).
Toomey 2008	OBSERVED SERVER BEHAVIOUR Pseudo‐intoxicated purchase attempts Total of purchase attempts in all participating establishments that were successful before the intervention and during two post‐intervention follow‐ups. One purchase attempt made in each establishment. Intervention                Control Baseline                    81/122 (66.4%)         68/109 (62.3%) 1st follow‐up              62 /111(55.9%)         67/105 (63.8%) 2nd follow‐up             73/111 (65.8%)          78/106 (73.58) Authors "observed no significant differences at follow‐up in reported policies/practices across establishments".
Wallin 2002	OBSERVED SERVER BEHAVIOUR 1. Refusal rates to intoxicated patrons (Data in paper published 2003.) 55% in the experimental sites which had received training, 48% in intervention sites yet to be receive training, and 38% in the control area. The authors reported that this was not significant. No further details presented. 2. Successful attempts to buy a drink by patrons who were over 18 but appeared to be under 18. (Data in paper published 2004.) Intervention Control Baseline (1996) 129/307 (42%) 57/146 (39%)  1st follow‐up (1998) 57/146 (39%) 46/106 (43%) 2nd follow‐up (2001) 37/118 (31%) 41/120 (34%) The authors reported that differences between intervention and control groups were not significant. No further details presented