Mandatory Responsible Beverage Service Training in California Associated With Higher Refusals of Service to Apparently Intoxicated Patrons

Abstract

Objective:

The California (US) Responsible Beverage Service (RBS) Training Act mandated training for all alcohol servers by 2023 giving us a rare opportunity to see how a statewide initiative might influence server behavior.

Method:

In 2022, 300 licensed on-premises establishments were sampled in nine counties representing the San Francisco Bay Area. From July 2022 to January 2023, “pseudopatron” (PP) and observer teams visited each establishment, and PPs attempted to buy alcohol while displaying obvious signs of intoxication. The outcome of each purchase attempt and characteristics of establishments, servers, PPs, and month, day, and time were recorded. In 2024, the same premises were visited as a follow up.

Results:

At baseline, 19.6% of the outlets refused service to the PP. At follow up, the refusal rate increased significantly to 39.2% in a regression model controlling for establishment, server and PP characteristics (OR=4.51, p<.01). A typology of outlets (e.g., sports bar; casual restaurant; fine dining) did not show any association with refusal rate except that fine dining (upscale) places refused more often than other types of establishments. Observer rating of how obvious the PP’s apparent intoxication was strongly associated with refusals (OR = 11.64, p<.01).

Conclusions:

This study suggests that there was substantial improvement in alcohol over-service refusals two years after the mandatory RBS training law went into effect.

Introduction

Alcohol service to obviously intoxicated patrons in on-premises establishments remains a public health concern as it is associated with an elevated risk of problems such as driving while intoxicated, interpersonal violence, and unintentional injuries. One approach to reducing alcohol overservice is providing training for servers to improve their skills in recognizing intoxicated patrons and effectively refusing service.

Responsible Beverage Service (RBS) refers to the steps that servers of alcoholic beverages can take to reduce the chances that their patrons (or guests) become intoxicated, or failing that, to intervene to reduce the risk of subsequent harm to the patron or others. This preventive intervention is perhaps unique in that it occurs just when the risk arises (i.e., while the patron is drinking or has become impaired), and its invocation does not depend on the drinker’s judgment.

The intervention is a variant of restricting access to alcohol, in this case, refusing access of an alcoholic beverage to a patron who is intoxicated. In most U.S. states, it is illegal to serve “obviously intoxicated” patrons and in every state, to patrons under 21 years of age. Compliance with the over-service law, however, is highly variable and usually quite low. Toomey and colleagues (2016) reported alcohol service refusal rates of about 18% to apparently intoxicated patrons among licensed outlets in one Midwest metropolitan area. Woodall and colleagues (2018) reported higher refusal rates of about 35% in New Mexico, where RBS training is mandatory for servers. More recent studies have reported disparate refusal rates, including 6.8% in California, 33.9% in New Mexico and 27.1% in Washington (refusals in both of two visits) (Buller, et.al., 2024). The baseline refusal rate for the present study in California was 21% (Saltz, et.al., 2024). It is clear that improving these refusal rates has the potential for significantly reducing driving while intoxicated and related road injuries and fatalities, but the question is how to achieve it.

A common approach is to provide server training to improve responsible alcohol service. A typical curriculum would include a review of state laws and regulations pertaining to service to minors or “obviously intoxicated” patrons, a brief overview of alcohol’s effects on the drinker, recognizing signs of intoxication, alcohol-related harms, and strategies for refusing service to intoxicated patrons. State or local agencies may add to these topics. The training may take 2 to 4 hours or more, and in-person training can include role-playing sessions. Online training can mimic role-playing by offering scenarios for the trainee to respond to.

The efficacy of training as a mechanism is uncertain. Ker and Chinnock (2008) find that there is “no reliable evidence that interventions in the alcohol server setting are effective in reducing injury” (ibid, page 13). They also note the general weakness in that literature and recommend greater use of randomized trials. Jones and colleagues (2011) find a bit more promise in their review of RBS studies, albeit with the same concern with the quality of the studies. The Community Preventive Services Task Force likewise found insufficient evidence to determine the effectiveness of RBS training programs, “…although reviewed studies generally showed positive results for measured outcomes...” (CDC:2010). A more recent review concluded that RBS can be effective (Fell, et.al, 2024). Given the mixed results of these studies, some have argued that continued research of RBS training is unprofitable (Buvik & Rossow, 2017)

Yet some studies have found significant positive effects, suggesting that it may not be the underlying strategy that is found wanting, but rather the quality of the training and the nature of its implementation. One of the earliest evaluations of server intervention - the Navy Server Study (Saltz, 1985, 1987; Hennessy and Saltz, 1989) - involved training staff and management over several hours accompanied by changes in serving practice to facilitate staff monitoring of patron consumption. The prevalence of legal intoxication (imputed from direct observation and self-report) was reduced by half (from 33% to 15% among men). The STAD project in Stockholm increased refusals of service to seemingly intoxicated pseudopatrons from 5% to 70% and reduced assaults and threats by 29% (Wallin, et.al., 2003; 2005), an effect that has been sustained over 20 years (Elgán, Andréasson, & Gripenberg, 2024) as have reductions in service to underaged youth (Feltmann, et.al., 2025). It also appears that the STAD intervention, when disseminated throughout Sweden, reduced violence more widely (Brännström , et.al., 2016). A replication in Oslo could not reproduce such results, although its implementation differed from the STAD project (Rossow & Baklien, 2010). A key study looked at refusal rates before and after implementing three different community RBS interventions. One site had RBS training alone; a second had enforcement alone, and a third had both. Refusal rates increased from 20.9% at baseline to 42.1%, 68.8%, and 74% respectively (Quigg, et.al., 2022). The results support the value of a multicomponent intervention but also demonstrate that enforcement can be effective even without training, as earlier reported by McKnight and Streff (1994). In New Mexico, a randomized trial was conducted of an online RBS training against usual and customary training. Refusals rose from about 35% at baseline to 67% immediately after training and was sustained one year later (Woodall, et.al., 2018). Finally, when Oregon became the first state to mandate RBS training, alcohol-involved motor vehicle crashes dropped by 23% net of other influences (Holder & Wagenaar, 1994). This was a surprisingly large effect given that not all servers had been trained, and training was the only element in the initiative (i.e., no enforcement).

Currently in the U.S., 23 states mandate RBS training for on-premises servers in some capacity. Nine require it for licensees, 21 states for managers, and 20 for servers (APIS, 2025). States may also or instead provide some form of incentive for RBS training (26 states). Such incentives may include its use in a possible liability lawsuit, mitigation of penalties, or discounts for training (APIS, 2025).

The Responsible Beverage Service (RBS) Training Program, mandated by California Assembly Bills 1221, required the Department of Alcoholic Beverage Control (ABC) to develop and implement an RBS training curriculum by January 1, 2020, and required all alcohol servers and managers to complete an approved training course by July 2021. The RBS program was officially implemented on July 1, 2022, after a one-year delay in light of the COVID pandemic, with a grace period extending compliance to August 31st. The state’s curriculum comprised responsible beverage service, state laws, regulations, and alcohol’s effects on the body. Trainees are required to complete training by an ABC-approved RBS training provider and pass an exam administered by the ABC. Licensees are responsible for ensuring their employees are certified but are not required to maintain compliance records for servers they employ. As of August 31, 2022, the ABC reported that over 300,000 individuals successfully completed training and passed the state RBS exam (California ABC, 2022). There are no data on the number of servers and managers subject to the law, but the ABC notes that in May 2021, the US Bureau of Labor Statistics estimated there were 210,000 bartenders and waiters and waitresses in California.

For this paper, the California training mandate provides a unique opportunity to address the question of whether such training alone can have an effect on alcohol servers’ compliance with laws forbidding service to obviously-intoxicated patrons. We compared the rate of refusing service to “obviously intoxicated” pseudo-patrons from just before the deadline for training, to the refusal rate two years later.

Methods

Sample of On-premises Licensed Outlets

A sample of 300 licensees was selected from all those with either beer and wine or general (including distilled spirits) licenses in 9 California counties that constitute the San Francisco Bay Area. These include metropolitan areas (e.g., San Francisco, San Jose), rural areas of Napa and Sonoma counties, and suburban communities in Marin, the East Bay and the San Francisco Peninsula. Altogether, these counties comprise nearly 8 million people and approximately 10,600 outlets licensed to sell alcohol for consumption on premises.

A multi-stage sample of 300 outlets was created with an initial random sample of 60 “seed” outlets from publicly available online lists of licensees published by the state Department of Alcoholic Beverage Control (ABC). Four more outlets were randomly selected from within the same zip code areas as the 60 “seeds.” This was to make data collection more efficient by reducing travel time between outlets. Anticipating the possibility that some outlets would be permanently closed, field teams were given a short list of randomly-selected “replacements” from the same zip code that they could use when necessary.

Of the 300 establishments assessed in 2022–23 (round 1), 255 were still operating with an active license and had complete data for all study variables. Specifically, 272 were still operating and re-surveyed in 2024 (round 2) (Saltz, et.al., 2024). Of these establishments, 262 still had an active on-sale license from the California ABC (10 did not), and of these, 255 had complete data for all study variables.

Pseudo-patron (PP) Protocol and Data Collection

Eight young adults at baseline (3 male) and 12 at follow up (6 male) were hired and trained to be alternating pseudo-patron (PP) buyers and observers. Several modes of recruitment methods were used, in all cases requiring acting experience. Research confederates were trained to perform the pseudopatron (PP) protocol (Buller et al., 2020; Woodall et al., 2018). In a 4-hour training, research confederates learned the purpose of the assessment, procedures for observer and PP (including ways to respond to establishment staff), and how to complete the observer and PP forms. Confederates were trained to enact signs of a high level of intoxication (see Table 1). To obtain realistic displays, PP were instructed on how alcohol affects the body and changes physical behavior. They were trained to display cues at levels that were obvious but not so obvious as to be unnatural or disruptive. They practiced PP and observer procedures and had intoxication displays videotaped, receiving corrective feedback when needed, although this was uncommon, as simulating intoxication was simple and easily recognized by others during the training. About 4 people declined to participate in the study after completing the training for various reasons, including reluctance to perform in the drinking venues. Training included practice visits to licensed on-premises establishments near the study office in Berkeley.

Table 1.

Study variables

Alcohol service refusal	A dichotomous variable was created to indicate whether or not an alcoholic beverage was served to PPs (0 = alcohol service, 1 = alcohol service refusal).
License type	Establishments were classified by license type (beer/wine or general) provided by a public database maintained by the California Department of Alcoholic Beverage Control.
Establishment types	Observers recorded whether the establishment was a restaurant, restaurant with a separate bar, or just a bar. Two dummy variables were created to represent bars and restaurants with a bar area with restaurant as the referent category. Establishments were also characterized using 10 categories from the National Restaurant Association (e.g., sports bar, brewpub, nightclub, theme bar, hotel bar, casual restaurant with table service, fast casual restaurant with ordering from a counter, upscale/fine dining restaurant). These variables were dummy coded, but the characteristics were not mutually exclusive.
Activity in establishments	Observers rated how busy each establishment was on 4-point ordinal scale (1 = Quiet to 4 = Very busy) and estimated the number of staff and patrons.
Warning signs	Observers recorded whether there were obvious warning signs indicating that (1) anyone under 21 years of age would not be allowed in the establishment or would not be served alcohol, and (2) alcohol was a risk for cancer and for someone who is pregnant. Both of these variables were dummy coded (0 = No, 1 = Yes).
Server characteristics	PP buyers recorded the type of staff engaged with them during the purchase attempts (e.g., bartender, table server, manager, bouncer/security). Because the majority of staff that engaged were bartenders or table servers, these two variables were dummy coded and other types of staff combined were the referent group. PP buyers also recorded the server’s approximate age, gender, ethnicity, and race.
Pseudo-patron characteristics	PP buyers’ gender was recorded. All PP buyers were young adults (21–29 years old). This reflected the availability and interest of actors in applying for temporary and part-time work. The number of PP buyers in different ethnic/racial categories was too small to be included in analyses.
Display of intoxication	Observers recorded how obvious the PP buyers’ display of intoxication was on a 5-point scale (1 = Not at all obvious to 5 = Very obvious). Because of the skewed distribution and no PPs classified as having a ‘not at all obvious’ display of intoxication, this variable was dichotomized (0 = Slightly/somewhat obvious, 1 = Obvious/very obvious).
Signs of intoxication	Observers also recorded whether PP buyers displayed seven different signs of intoxication (e.g., slurred speech, swaying, stumbling, lurching, dropped ID or money, forgetting things, repeating things). These indicators were summed to create an overall measure of signs of intoxication.
Time of data collection	The date and time of each establishment visit was recorded. From the dates, we created separate variables for month, day and hour of each visit.

During data collection, PP and observer teams were sent to a group of 5 outlets on Thursday, Friday, and Saturday nights. A PP buyer would feign intoxication by displaying obvious intoxication cues and record whether they were served a drink (beer) when it is illegal to do so. This technique was used in over 2500 buying attempts by McKnight (1993), as well as adopted by others who have evaluated RBS programs in the U.S., Canada, and European countries (e.g., Gosselt et al., 2013; Quigg et al., 2022; Toomey et al., 2005, 2016;Elgán et al., 2021). PP buyer assessments present the server with the most overt situation in which service is supposed to be refused (i.e., when a patron is displaying clear signs of intoxication).

The observer entered the premise first, ordered a non-alcoholic beverage and recorded characteristics of the premise environment (e.g., type of establishment, approximate number of patrons and staff). The PP buyer entered the establishment, exhibiting the signs of intoxication they were trained to present and attempted to order and purchase a standard beer. The observer recorded the nature of the interaction between PP buyer and server and descriptive data on the establishment. The observer also recorded the extent to which the PP buyer exhibited the trained behaviors of intoxication and details of the interaction with the server after s/he exited the establishment. The PP buyer recorded whether service was made as well as whether there were any comments expressing concern or worries about the apparent condition of the PP, whether food was offered, or whether the server made a joke about it (a rare occurrence). The PP also noted whether they were refused entry, whether any other staff or manager were brought into the interaction, and whether any other patrons engaged in the interaction. The PP buyer and observer also recorded the server’s estimated age, ethnicity and gender. Additionally, the observer recorded date and time of the visit, whether the place was a restaurant, restaurant with a separate bar, or just a bar. They recorded establishment types using categories from the National Restaurant Association (e.g., casual with table service, fast casual ordering from a counter, fine dining). They rated how busy the business was, estimated the number of staff and patrons and reported how obvious the PP’s display of intoxication was. All of these data were input into data collection forms on tablet computers by observers and PP buyers to facilitate data entry. Establishment-level PP buyer and observer data were combined for subsequent analyses.

Study variables are shown in Table 1. In addition, an owner/manager survey was conducted at a convenience sample of 104 establishments across California in 2024. The sample was recruited through postcard mailings with a link to an online survey and through in-person visits at establishments in the SF Bay Area. The survey included questions about whether the owner/manager was familiar with the alcohol server training law, how many staff had received RBS training, staff turnover, the owner/manger’s attitudes and beliefs about RBS training, and establishment characteristics.

Data Analysis

To assess possible sample attrition bias, we compared the round 1 alcohol service refusal rates and characteristics of establishments that were and were not included in both survey rounds, along with characteristics of servers and PP buyers (chi-square and t-tests). We also compared these variables in survey rounds 1 vs. 2. Mixed effects logistic regression analysis was then conducted to examine the likelihood of alcohol service refusal in survey round 2 vs. round 1 when controlling for other study variables. We also examine possible differences in change in the likelihood of service refusal among pseudopatrons rated by observers as displaying “obvious or very obvious” versus “slightly or somewhat obvious” signs of intoxication by including a survey round × display of intoxication term in the regression model. The regression analysis was conducted in Stata version 18 and accounted for repeated establishment observations and used robust standard errors (StataCorp, 2023). The statistical significance threshold was set at p=0.05 (two-tailed). To assess multicollinearity, we used the collinearity diagnostic procedure in Stata with all study variables, which generates a variance inflation factor (VIF) for each variable. VIFs ranged from 1.06 to 2.49 (mean = 1.45), indicating low multicollinearity.

Additional analyses were conducted to examine possible server gender, server age, and PP gender differences in the effect of the alcohol server training law on service refusal by including corresponding interaction terms (e.g., survey round × server gender) in the regression model.

Descriptive statistics from the owner/manager survey were examined to assess compliance with the alcohol server training law and owner/manager attitudes and beliefs about RBS training.

Results

Sample Attrition Analyses

As shown in Supplemental Table 1, there was no statistically significant difference in alcohol service rate in round 1 between establishments not included or included in the round 2 assessment, although the percentage of refusal did appear higher in those establishments not included than included in round 2 (31.1% v. 19.6%). The two groups were similar with respect to other establishment characteristics and server characteristics and PP buyer characteristics except the percentage of PP buyers who were Asian was lower among establishments that were not included in both survey rounds (17.8%) compared to those that were included (32.9%). Additionally, the two groups were similar with respect to month, day and hours of pseudo-patron visits, except that a lower percentage of establishments not included in both survey rounds were visited during the 5–7 pm evening period (11.1%) compared to those that were included in both rounds (24.7%).

Survey Round Comparisons

Results in Table 2 indicate a significantly higher alcohol service refusal rate in survey round 2 (39.2%) compared to round 1 (19.6%). Although most establishment characteristics were similar in both rounds, there were some differences. A significantly lower percentage of establishments in round 2 had casual dining with table service (43.1%) compared to round 1 (58.0%). There was a significantly lower level of observed establishment activity and a significantly lower number of patrons in establishments in round 2 compared to round 1, but a significantly higher percentage of establishments had a warning sign regarding the minimum legal drinking age in round 2 (22.4%) compared to round 1 (7.1%). Some server race group characteristics (% white, other and unknown race) differed significantly in round 2 vs. round 1. Pseudo-patrons were similar with respect to gender and observed display of intoxication in each round, though significantly fewer signs of intoxication were observed in round 2. There were also significant differences in the percentage of establishments visited during each month during rounds 1 and 2, but no significant differences with respect to days of the week and evening hours.

Table 2.

Sample characteristics by pseudo-patron survey round (N=255), percent or mean (SD)

	Survey round
Variable	1	2	χ2(df) or t(df)	p
Alcohol service refusal	19.6	39.2	23.61(1)	.001
Establishment characteristics
Beer/wine license	52.5	53.7	0.07(1)	.79
General license	47.5	46.3	0.07(1)	.79
Restaurant with no bar area	44.3	40.8	0.65(1)	.42
Restaurant with bar area	44.7	46.3	0.13(1)	.72
Bar	11.0	12.9	0.46(1)	.49
Sports bar	10.2	9.4	0.09(1)	.77
Theme bar/restaurant	18.0	20.4	0.45(1)	.50
Brewpub	3.5	3.5	0.00(1)	1.00
Winery/wine bar	2.7	2.4	0.08(1)	.78
Dive bar	8.2	5.1	2.02(1)	.15
Hotel bar	3.1	2.4	0.29(1)	.59
Music bar/nightclub	4.3	1.6	3.37(1)	.06
Upscale/fine dining	16.1	14.9	0.13(1)	.71
Fast casual/fast food	24.3	18.4	2.62(1)	.10
Casual dining w/table service	58.0	43.1	11.33(1)	.001
Establishment activity	2.2 (1.1)	2.0 (0.9)	2.08(508)	.04
Number of patrons	19.0 (21.1)	14.0 (20.4)	2.73(508)	.007
Number of staff	4.5 (3.9)	4.3 (2.9)	0.62(508)	.53
Warning sign – under 21	7.1	22.4	23.77(1)	.001
Warning sign – pregnancy/cancer	36.9	32.2	1.25(1)	.26
Server characteristics
Bartender	41.2	38.0	0.52(1)	.47
Server	42.4	46.7	0.96(1)	.32
Age	35.8 (11.2)	35.2 (9.7)	0.65(508)	.51
Male	52.5	54.1	0.13(1)	.72
Hispanic/Latino	32.2	32.2	0.00(1)	1.00
Asian	32.9	31.0	0.22(1)	.63
Black	3.1	0.8	3.67(1)	.05
White	32.9	21.2	8.94(1)	.003
Other race	25.1	17.3	4.69(1)	.03
Unknown race	5.9	29.8	49.77(1)	.001
Pseudo-patron characteristics
Female	55.7	50.6	1.33(1)	.25
Obvious/very obvious display of intoxication	74.1	67.8	2.44(1)	.12
Signs of intoxication	3.8 (1.1)	3.4 (1.2)	3.76(508)	.001
Month
May	0.0	12.2	33.00(1)	.001
June	0.0	45.9	151.83(1)	.001
July	20.4	33.7	11.48(1)	.001
August	42.4	8.2	78.54(1)	.001
September	34.9	0.0	107.81(1)	.001
October-January	2.4	0.0	6.07(1)	.01
Day
Thursday	31.8	31.0	0.04(1)	.85
Friday	36.1	35.3	0.03(1)	.85
Saturday	32.2	33.7	0.14(1)	.71
Hour
5–7 pm	24.7	32.5	3.84(1)	.05
7–8 pm	42.7	38.4	0.98(1)	.32
8–10 pm	32.5	29.0	0.74(1)	.39

Note: Significance tests are based on comparisons of percentages or means for study variables among establishments in survey round 1 vs. round 2.

Regression Analysis

Results of mixed effects logistic regression analysis (Table 3) indicate that establishments were more than four times as likely to refuse alcohol service to pseudo-patrons in survey round 2 compared to round 1 when controlling for establishment, server, and pseudo-patron characteristics and time periods when establishments were visited. Alcohol service refusal was more likely to occur at upscale/fine dining establishments, when servers were Asian American, and when pseudo-patrons were observed to have an obvious or very obvious display of intoxication. None of the other study variables was associated with the likelihood of alcohol service refusal.

Table 3.

Results of logistic regression analyses predicting the likelihood of alcohol service refusal, odds ratio (95% confidence interval)

Variable	OR (95% CI)	z	p
Survey round 2	4.51 (2.17, 9.35)	4.05	<.001
Establishment characteristics
Cluster - general license, bar, more staff and patrons, bartender	0.87 (0.49, 1.56)	−0.45	.65
Upscale/fine dining	2.37 (1.22, 4.62)	2.54	.01
Casual dining with table service	0.61 (0.35, 1.07)	−1.71	.09
Theme bar/restaurant	0.64 (0.32, 1.26)	−1.30	.19
Establishment activity	0.96 (0.73, 1.26)	−0.28	.78
Warning sign – under 21 not allowed	1.72 (1.01, 2.91)	2.01	.04
Warning sign – pregnancy/cancer	0.87 (0.42, 1.80)	−0.37	.71
Server characteristics
Age	1.01 (0.99, 1.03)	0.95	.34
Male	0.88 (0.53, 1.45)	−0.50	.62
Hispanic/Latino	0.61 (0.28, 1.33)	−1.24	.21
Asian	2.28 (1.06, 4.92)	2.11	.03
Black	3.11 (0.36, 26.86)	1.04	.30
Other race	2.08 (0.83, 5.23)	1.57	.12
Unknown race	1.92 (0.77, 4.82)	1.41	.16
Pseudo-patron characteristics
Female	1.31 (0.77, 2.22)	1.01	.31
Obvious/very obvious display of intoxication	11.64 (4.73, 28.65)	5.34	<.001
Signs of intoxication	0.99 (0.79, 1.24)	−0.10	.92
Month
July	1.50 (0.84, 2.68)	1.38	.17
August	1.68 (0.83, 3.39)	1.45	.15
Day
Friday	0.91 (0.50, 1.65)	−0.32	.75
Saturday	0.98 (0.55, 1.76)	−0.05	.96
Hour
7–8 pm	0.98 (0.55, 1.73)	−0.07	.94
8–10 pm	0.81 (0.45, 1.46)	−0.71	.48

Note: Referent categories include: survey round 1, establishments with a beer/wine license, restaurants, fewer staff and patrons, and table servers (Cluster); white servers (Server characteristics); May/June/Oct.-Jan. (Month); Thursday (Day); and 5–7 pm (Hour).

^*p<.05

^**p<.01

An additional regression model including the survey round × display of intoxication term indicated that the interaction term was not statistically significant [odds ratio (95% CI) = 0.77 (0.15, 4.04), p = .76]. We note, however, that among pseudopatrons rated as having a “slightly or somewhat obvious” display of intoxication, the service refusal rate increased four-fold, from 3.4% to 12.3%, while the service refusal rate doubled for pseudopatrons rated as having an “obvious or very obvious” display of intoxication, from 24.9% to 52.4%. Additional regression analyses were conducted with a variable representing the predicted probability of attrition from the initial sample of 300 establishments. Results of these analyses were very similar to those obtained without this adjustment for attrition.

Additional regression analyses did not indicate significant interactions between survey round and server gender or server age. However, there was a significant interaction between survey round and PP gender [OR (95% CI) = 0.38 (0.18, 0.81), z=−2.50, p=.01]. Further analysis indicated a stronger association between survey round and the likelihood of service refusal among male PPs [OR (95% CI) = 10.89 (3.51, 33.74), z=4.14, p<.001] compared to female PPs [OR (95% CI) = 4.15 (1.40, 12.22), z=2.58, p=.01]. These results are illustrated in Figure 1.

Figure 1.

Percentage of establishments that refused alcohol service to pseudo-intoxicated patrons by survey round and gender

Establishment Owner/Manager Survey

Results in Supplemental Table 2 indicate that the majority of owners/managers were at least somewhat familiar with the alcohol server training law (87.5%) and most/all servers (63.4%) and owners/managers (69.2%) had received state-approved RBS training (by manager report). Staff turnover was low, with about 85% of the managers reporting no or few staff changes. The majority of owners/managers expressed favorable attitudes towards RBS training (e.g., reduces chance of legal problems and helps management), and 83.6% believed that an establishment would receive a warning or citation by the California ABC or local police if they served an obviously intoxicated customer. However, 47.1% agreed or strongly agreed that customers should bear the responsibility of drinking alcoholic beverages, not servers.

Discussion

Research on the effectiveness of responsible alcohol server training programs has yielded inconclusive results with respect to refusal of service to obviously intoxicated patrons. This study examined possible effects of the California alcohol server training law on refusal of alcohol service to pseudo-intoxicated patrons in a large sample of licensed on-premises establishments. Findings suggest that the law was effective as the alcohol service refusal rate increased from 19.6% to 39.2% over a two-year period. Controlling for a variety of factors the difference in refusals amounted to being more than four times as likely in follow-up. Larger, though non-significant increase in the likelihood of refusals when pseudopatrons were less “obvious” than for “very obvious” pseudopatrons, suggests servers were paying closer attention to the patrons. These findings parallel those state-level evaluations in Oregon (Holder & Wagenaar, 1994) and New Mexico (Woodall, et.al., 2018) showing positive effects of RBS training. This is a welcome achievement for the new law, but it still means that obviously intoxicated pseudopatrons were served in well over half of their visits to licensed alcohol outlets.

This policy study is by design limited in its ability to address server behavior more generally. Nevertheless, there are points of comparison with prior research. First, where others have found that women are less likely to be refused service (e.g., Buvik & Baklien, 2016), we found women to be refused more often than men at baseline, with no difference at follow up, suggesting that refusals by gender may be fluid, at least in the U.S.

Second, we found no association of refusal rates with type of business, with the exception of “fine dining” restaurants. One might expect that bars would differ from restaurants, but in these data, that was not the case. Several studies have linked low refusal rates to “aversive environments” (e.g., noisy, crowded, intoxicated patrons) (Roberts, 2009; Buvik & Rossow, 2015). Noise was not recorded here, but degree of “busyness” or “crowdedness” had no bearing on refusal rates.

Several studies involved interviews with servers themselves (Burns, et.al., 2003; Reiling & Nusbaumer, 2006; Buvik, 2013; Gosselt, et.al., 2013; Leo, 2013; Buvik & Baklien, 2016; Zafer, et.al., 2018). A common theme was servers citing a “drinking culture” that sees intervention as illegitimate or may be reflected in the servers’ own attitude about serving and not intervening. Such a culture is likely to differ from one society to another and may be a reason why evaluations of RBS find different or no effects. Perhaps in the U.S., with its history of dram shop law and civil liability, there is more (if limited) support for the concept of RBS than might be found in, say, Nordic societies.

If true, should a drinking culture be considered a fixed, limiting factor for RBS or can an RBS intervention be in part a strategy for ameliorating the effect of such a culture? One function of regulation and enforcement is to communicate societal norms, including where such regulation may represent some change in expectations. Wholesale changes in law is unlikely, but new regulations and enforcement of existing ones can nudge societal norms about alcohol.

The success of the STAD intervention, and the three-condition community study reported by Quigg and colleagues (2022) form a basis for recommendations to implement community-level multicomponent programs (Hughes, et.al., 2010; Graham, et.al., 2014; Quigg, et.al., 2022; Zafer, et.al., 2018; Saltz, 2025) that often feature an enforcement component, or to create regulation where it might not exist.

Perhaps the increase in refusals reported here is the best one could expect from a mandated training without a more comprehensive intervention. In the context of the U.S., server training could be supplemented by a high-visibility enforcement campaign as per best practice of deterrence (Zeng, et.al., 2025). This could include visits by plainclothes law enforcement officers, follow up visits based on where those convicted of driving under the influence had been drinking, a coordinated patron/public awareness campaign, and publicizing cases where a business was found liable for serving an intoxicated patron as is currently done for serving minors. Direct enforcement of serving laws is challenging, with no analog of underage “decoys” where service to them is clearly breaking the law. This is likely the reason why, in the U.S., enforcement of sales to minors is much more likely than enforcement of overservice (Lenk, et.al., 2023).

RBS training implicitly assumes that poor serving practices stem from either ignorance of the law, inability to recognize signs of intoxication, or ineffective ways to refuse service. The California law’s effect may then be attributed to educating servers about their responsibilities and conferring skills to practice them. These issues point to the need for process evaluation to clarify what is essential for a successful RBS intervention.

We have seen, in the research literature, much larger effects for those RBS initiatives that are community-wide and where RBS training is just one element of a multicomponent intervention that also included monitoring and enforcement of serving laws (Quigg, et.al., 2022; Burton, et.al., 2017). This suggests that the California RBS law could be supplemented with even greater effect. Future research should focus on evaluating multi-component interventions, including replications of the STAD project, and with attention to rigorous evaluation designs.

Limitations

There are several limitations in this study that might affect its validity. First, the sample of alcohol outlets was limited to the San Francisco Bay Area and thus may not reflect serving behavior in other parts of the state. The simple pretest/posttest design with no comparison group cannot rule out the possibility of some event or trend that could have caused the same result. This is a common weakness of research that tries to take advantage of a “natural experiment” and is thus outside the control of the investigator. Another study using the same pseudopatron protocols in New Mexico was being conducted over roughly the same time as this one (2022 and 2024 follow up) and found no major change in refusals (52.1% to 55.9% baseline to follow up, n=59 visits) (Buller, et.al., 2024; personal communication), suggesting there was no national or regional factor that might account for the increase observed in California.

Sample attrition may have introduced bias into our results. There was a higher round 1 refusal rate for the establishments not included in both rounds (31.1%. vs. 19.6%), raising the possibility that our finding regarding the increase in the refusal rate from round 1 to 2 may be an underestimate of the actual increase. However, any attrition bias was likely small because (a) the number of lost establishments was very small and (b) the difference in alcohol service rates between those included or not included in round 2 was statistically non-significant. In addition, lack of adjustment for multiple comparisons in analyses may have resulted in results biased towards statistical significance.

The server’s behavior was in response to pseudopatrons acting intoxicated. While trained in ways to display intoxication, there may have been some reason their displays were more obvious in the post-test visits, although the observers’ ratings of “obviousness” were no different pre- and post-training.

We had no way of knowing whether the server in each encounter had actually been trained in RBS, though California’s Department of Alcohol Beverage Control reported that the number of those who successfully completed a course were close to their estimate of the total number of servers (CA ABC, 2022). If true, it seemed unlikely that the servers were not trained. Finally, there is no way to know how the pandemic may have affected the study.

Despite these limitations, the results of this study do suggest a notable impact of the California Alcohol Server Training Act. Improvements in refusal of service to apparently intoxicated patrons reduces the likelihood of intoxicated driving, other socially harmful behavior, and improves community safety. Further efforts can be considered to bolster these effects.