Randomized Trial Testing an Online Responsible Vendor Training in Recreational Marijuana Stores in the United States

Abstract

Objective:

An online training in responsible marijuana vendor (RMV) practices was evaluated for effects on compliance with ID checking regulations.

Method:

A random sample of state-licensed recreational marijuana stores (n = 175) in Colorado and Washington State was selected in 2016–2017 and was enrolled in a randomized pretest-posttest controlled design. After baseline assessment, 75 stores were randomly assigned to a usual and customary training control group, stratified by state and region. The remaining stores (n = 100) were invited to use the RMV training. Stores were posttested at 3 and 9 months postrandomization. The primary outcome was refusal of sale measured with pseudo-underage patrons who attempted to enter stores and purchase cannabis without a state-approved ID.

Results:

There was no difference by treatment group in refusal of pseudo-underage patron buyers (baseline: 92.5% intervention vs. 94.7% control; 3-month post-test 94.8% vs. 97.5%; 9-month posttest 97.5% vs. 97.1%, p = .286 [one tailed, adjusted for covariates]). The use of training increased refusals at store entry (trained: 65.9% baseline 82.5%, 3 months 79.9%, 9 months; not trained: 82.6%, 83.1%, 84.5%, p = .020 [two tailed, adjusted for covariates]). This difference was especially evident in Washington State (trained: 40.3%, 65.1%, 60.4%; not trained: 57.9%, 68.5%, 72.3%) but not in Colorado (trained: 95.2%, 101.0%, 101.4; not trained: 95.7%, 98.6%, 99.2%, p = .033 [two tailed, adjusted for covariates]).

Conclusions:

When used by store personnel, online RMV training increased refusal of buyers who appeared young and did not provide a state-approved ID. However, it did not improve refusal rates overall. Stores that had lower refusals at baseline and used the training may have benefited from it.

Currently, recreational marijuana is sold in 10 U.S. states at state-licensed retail stores to individuals age 21 or older who provide a valid state-approved identification (ID) (Berke & Gould, 2020), and voters in four more states legalized sales in the November 2020 election (Lopez, 2020). The U.S. Congress may vote in the near future on a bill removing marijuana from the Controlled Substances Act (Demko et al., 2020; Ferris & Fertig, 2020). Recreational marijuana sales are also legal in Canada (Austen, 2017) and Uruguay (Maybin, 2019).

Marijuana availability has promoted the initiation of use in the past (Miech et al., 2015; Swahn & Hammig, 2000; Swaim, 2003; von Sydow et al., 2002), leading to concerns over legalization (Damrongplasit et al., 2010; Joffe & Yancy, 2004; MacCoun & Reuter, 2001; Pacula et al., 2004). At the outset of sales, the U.S. Department of Justice (DOJ) notified states that they must create strong controls to prevent distribution to youth (Cole, 2013) (although this memorandum was rescinded, prosecutors are still adhering to it; Firestone, 2020). In many states, alcohol regulators, tasked with regulating recreational marijuana sales, promulgated rules regarding verifying customers’ age to prevent youth access.

Responsible beverage service (RBS) training is an intervention in more than 36 states that aims to prevent sales to minors (Ker & Chinnock, 2008; National Institute on Alcohol Abuse and Alcoholism [NIAAA], 2013). RBS has reduced outcomes indicative of illegal alcohol sales to minors (e.g., beer consumption; automobile crashes by drivers under 21) (Fell et al., 2016; Jones et al., 2011; Scherer et al., 2015). Presently, four states require responsible marijuana vendor (RMV) training for recreational marijuana store staff, and another state incentivizes it.

In this article, results of the first randomized trial to evaluate an RMV training in Colorado and Washington State are reported. Both states restrict sales of recreational marijuana to adults age 21 or older and require stores to verify age by checking a state-approved ID. Although both states do not allow individuals younger than age 21 to be on the premises, at the time of the trial, there was a key difference in who was liable for sanction—the underage individuals themselves in Washington State but not in Colorado, where only store personnel could be sanctioned. Although not specified in law, at baseline, Colorado store employees were likely to check IDs twice: before letting individuals enter the store and at the counter. In Washington State, personnel were likely to check IDs only once (Buller et al., 2019a). The trial tested the following hypothesis: Recreational marijuana stores that are randomly assigned to receive the RMV training will increase refusal rates of pseudo-underage buyers who do not present state-approved ID at 3- and 9-month follow-ups more than stores receiving usual and customary training.

Method

Store sample

Recreational marijuana stores (n = 175) were randomly selected in September and October 2016 from licensee lists from the Colorado Marijuana Enforcement Division (n = 85 stores) and the Washington State Liquor and Cannabis Board (n = 90 stores) by the project biostatistician. Some stores sold recreational marijuana exclusively, and others also sold medical marijuana. Stores were excluded if the license was inactive, suspended, or revoked or if any of their personnel participated in our formative research. Store selection was stratified by state and region (i.e., largest metropolitan area [Denver n = 40 and Seattle n = 38] vs. outlying areas with small cities and rural areas within a day’s driving distance of Denver [n = 45] and Seattle [n = 52]). A list of alternative stores was selected to replace closed stores before randomization. The sample was planned to achieve 0.85 statistical power to detect an increase of 9% in refusal rate at p < .05 (one tailed). Among intervention stores, employees who were invited to use the RMV training met the following inclusion criteria: owned or were employed at a licensed store, sold recreational cannabis products, and were age 21 or older (prescribed by state laws).

Trial design

The trial involved a pretest-posttest randomized controlled design. Beginning in October 2016, a baseline evaluation was conducted with all stores using a pseudo-underage patron protocol. Confederates age 21 or older who appeared young attempted to enter the store and purchase cannabis without providing a state-approved ID (Buller et al., 2016). Following baseline, 75 stores were randomly assigned to a usual and customary training (UCT) control group by the project biostatistician, using SAS random number generator stratified by state and region; these stores were not contacted except for follow-up assessments. The sampling universe was known in advance, so a predefined number of stores was used to select randomly from among strata. The other 100 intervention stores were contacted by research staff and were invited to use the RMV training. Store employees completed consent forms and pre- and post-training surveys in the training website (Buller et al., 2019b). All stores were posttested with pseudo-underage patron protocol approximately 3 and 9 months postrandomization. For each assessment, assessment teams visited each store twice (i.e., 6 total visits), at least 3 weeks apart. All stores were blind to pseudo-underage patron assessments, intervention store managers were aware of RMV training, and store managers were debriefed at the end of the trial. The trial procedures were approved by the Western Institutional Review Board, and a Certificate of Confidentiality was issued by the National Institutes of Health.

Train To Tend® responsible marijuana vendor training

The RMV training, named Train To Tend, was an interactive online training with five modules: (a) The Laws—current state regulations for recreational marijuana sales; (b) ID Checking—information and skills to check IDs and spot underage customers; (c) Health Effects—harms and benefits of marijuana (Colorado only); (d) Customer Service—methods of communicating with customers, recognizing intoxication, refusing sales, and advising customers on responsible use; and (e) Rules of the Trade—inventory tracking, product labeling, advertising, and store safety. Development followed the Criterion Referenced Instructional Framework for skillbased learning (Mager, 1962) and learning levels in Bloom’s taxonomy (Bloom et al., 1966). Instructional elements included “infogadgets,” a set of tabs with text and graphics on a single topic, and interactive activities and simulations to apply skills and receive feedback (Horton, 2006). The training operationalized behavioral modeling principles of Social Cognitive Theory (Bandura, 1986, 2004). To achieve user engagement, animated case studies modeling skills and comprehension checks (i.e., questions with feedback) were presented periodically. Trainees took four 10-item end-of-module quizzes and a 50-item final examination. State regulators reviewed the RMV training and set minimum “passing” scores (70% Colorado and 80% Washington State).

The ID Checking module contained seven learning elements. Trainees first were presented with five IDs and had 8 seconds each to determine whether they were valid for purchasing recreational marijuana. Next, common ways minors attempt to obtain recreational marijuana were described in an infogadget. Users then were taught about state-approved IDs and steps for checking IDs in an interactive activity. Trainees participated in an ID checking simulation, examining a variety of ID types and indicating whether each was valid for purchasing recreational marijuana, and they received feedback. Steps for checking medical marijuana registry cards, tips for verifying the age of customers, and ways for preventing third-party sales (i.e., someone buying recreational marijuana for a minor) were presented in three final infogadgets. A 10-item module quiz included checking five IDs and five multiple choice/true-false questions. Formative research to develop Train To Tend and pre- and post-training assessments are reported elsewhere (Buller et al., 2019b).

Pseudo-underage patron assessment

Compliance with ID checking regulations was measured by trained confederate buyers visiting stores and requesting to purchase an inexpensive cannabis product (i.e., a prerolled joint) without presenting a state-approved ID (Buller et al., 2016). Male and female legal-age individuals (≥21 years old) who looked young and were chosen for their acting experience performed the pseudo-underage patron assessments, blind to the study purpose and treatment. In one baseline assessment visit in Colorado, minors acted as buyers and presented their driver license showing that they were underage (since Washington State deemed it illegal for minors to enter stores, minor protocol was infeasible). This minor protocol did not change refusal rates (Buller et al., 2019a), so all subsequent assessments involved legal-age individuals not presenting IDs. Similar protocols were used to assess compliance with alcohol and tobacco sales regulations (McKnight, 1993; Woodall et al., 2018).

Two-person pseudo-underage patron teams of a buyer and an observer, supervised by a field supervisor, visited each store twice in each assessment round between 8:20 a.m. and 9:30 P.m. on weekdays and weekends. During training, each team visited at least two stores not in the sample, performed assessment, reviewed it with the supervisor, and were re-trained if needed. Confederates alternated acting as buyer and observer. Buyers and observers entered the store together (the observers always entered first, presenting a state driver license showing the observer was age-eligible to purchase recreational marijuana). The buyers said they did not have an ID with them, if asked. If refused entrance, the buyers exited. The observers, meanwhile, entered the store and observed its environment. If allowed entrance, the buyers approached the counter, asked to examine inexpensive pre-rolled joints, and attempted to purchase one. The observers positioned themselves to observe buyer–employee interchange and to note whether the employee agreed to sell cannabis to the buyer. As required by the federal funding agency, buyers did not purchase the cannabis product, saying they did not have enough money, to avoid using federal funds to buy marijuana. Team members completed data forms after exiting, recording whether entry was permitted and whether the store employee agreed to sell marijuana (yes vs. no). The observers recorded whether the buyers were asked to show ID and requested an inexpensive marijuana product. The buyers and observers recorded race/ethnicity, gender, and age of the store employee(s) (average interrater reliability, κ = .82). The supervisor checked forms for completeness.

Store characteristics

From store websites, project staff members recorded whether a store sold recreational marijuana or recreational and medical marijuana. Observers described the store environment (i.e., cleanliness, lighting, number of employees, and signage).

Statistical analysis

The primary outcome variable was refusal of sale, defined as refusal either at entry (i.e., not allowed into the store), after entry and before reaching the counter (i.e., on the floor), or at the purchase attempt. Refusal overall and refusal at each of these locations were tested separately. Stores were nested within states. Repeated-measures analysis with clustered design compared refusal of sale between treatment groups, assessment rounds, and two-way interactions, with state, type of store (recreational vs. recreational and medical), and region (metropolitan area vs. outlying area) being controlled for. One-tailed alpha criterion of p < .05 was planned a priori for hypothesis testing because (a) the hypothesized effect of training was directional, (b) we had no evidence that RMV training is harmful, and (c) RMV training would be recommended only if it had a positive impact. The effects of training uptake and two-way interactions on refusal of sale were investigated, along with moderation of treatment/training uptake effects by state. These latter effects were considered exploratory; therefore, a two-tailed alpha criterion of p < .05 was used. PROC MIXED in SAS 9.3 (SAS Institute Inc., Cary, NC) was used to fit the models. Models were run on pseudo-underage patron observations performed, without imputing missing observations from closed stores, so as not to artificially increase statistical power. We tested one model in an intent-to-treat procedure, imputing missing observations as “not refused,” and confirmed this.

Results

Profile of the store sample

Of the 175 stores selected, 95 (54.3%) sold only recreational marijuana and 80 (45.7%) sold both recreational and medical marijuana. No stores sold only medical marijuana. Observers recorded an average of 3.2 (SD = 1.7) personnel working in the stores. Of the personnel who engaged with the pseudo-underage patrons, 22.5% were security personnel/bouncers, whereas all others were owners, managers, or clerks. Personnel appeared to be young (about 30.7 years [SD = 8.0] of age) and were 37.4% female, 70.5% non-Hispanic White, 10.7% Hispanic White, and 18.8% some other race/ethnicity. The pseudo-underage patron buyers were 55.3% female and 63.0% non-Hispanic White, with a mean age of 23.3 years (SD = 2.4). A total of 14 stores were lost to follow-up because of business closure (see CONSORT diagram in Figure 1). Comparisons of community characteristics from the U.S. Census showed that stores lost were in communities with older populations than stores fully assessed (i.e., with a lower percentage of 18–24-year-olds from the U.S. Census [8.7% vs. 10.8%, p = .012, two tailed]). Conversely, compared with the UCT control group, stores in the intervention group were in communities with a younger population (i.e., with a higher percentage of 25–34-year-olds [17.2% vs. 15.5%, p = .038, two tailed] and a lower percentage of population 65 years or older [12.6% vs. 14.2%, p = .045, two tailed]).

Figure 1.

CONSORT flow diagram of trial. CO = Colorado; WA = Washington.

Uptake of Train To Tend RMV training

Of the 100 stores invited to use Train To Tend, owners/ managers at 42 stores (42.0%) agreed to use the program and had at least one individual complete it between June 2017 and January 2018 (i.e., completed all units in sequence with no time limit). In these 42 stores, 461 employees registered for the training, and 423 (91.8%) completed it. Trainees were 45.1% female, 73.3% non-Hispanic White, 5.3% Hispanic White, and 21.4% of other race/ethnicity; 48.9% were age 26 or younger (40.3% 27–39 years old; 10.8% 40 or older). Also, 35.6% had worked for less than 6 months in the marijuana industry, 18.9% 6 months to less than 1 year, 32.1% 1 to less than 3 years, and 13.4% 3 or more years. More than three quarters of trainees (76.8%) had received other training at their store (shadowing another employee [81.9%], from manager/owner [77.3%], in a group setting [46.5%], or from an outside trainer [26.5%]). The mean completion time was 155.7 minutes (SD = 66.4). Trainees felt that Train To Tend was highly usable (M = 75.4 [SD = 15.3]; scale range: 0–100), gave it a 4- or 5-star rating out of 5 stars (75.5%; M = 4.0 [SD = 0.9]), and would recommend it to other store personnel (81.9%).

Comparison of treatment groups

The hypothesis that recreational marijuana stores assigned to the intervention group would show higher refusal rates at follow-up than stores in the UCT control condition was not supported. The refusal rate of pseudo-underage patron buyers in the intervention group did not increase from baseline (92.5%) to either 3-month follow-up (94.8%) or 9-month follow-up (97.5%) over the refusal rate in the UCT control group (94.7%, 97.5%, and 97.1%, respectively; p = .286 [one tailed]), with state, type of store, and region being controlled for. There was also no statistically significant interaction by state (p = .221 [two tailed]).

Effect of training uptake

When looking only at stores within the intervention condition, we saw that training uptake, defined as at least one employee having completed the training at stores in the intervention group, had a significant impact on the refusal of pseudo-underage patron buyers at the store entrance (Table 1). The refusal rates at the entrance in the intervention stores that used Train To Tend increased from baseline (65.9%) to either 3-month follow-up (82.5%) or 9-month follow-up (79.9%) over the refusal rates in the intervention stores that did not use Train To Tend (82.6%, 83.1%, and 84.5%, respectively; Training Uptake × Assessment Round interaction p = .020 (two tailed), with state, type of store, and region being controlled for). This effect was not evident in the entire sample, with control stores classified as “not trained” (p = .162 [two tailed]). The same effect occurred when training uptake was defined as the percentage of employees in each store that completed Train To Tend (managers’ reported number of employees in each store). Especially when a majority of employees had completed the training, the refusal rate at the store entrance increased substantially from baseline in the 3-month follow-up (p = .004 [two tailed] for the intervention group only; p = .033 [two tailed] for the entire sample).

Table 1.

Adjusted estimated refusal rates by training uptake at various locations in recreational marijuana stores

Variable	Baseline	3-month follow-up	9-month follow-up
Training uptake—Any employee trained (entire sample n = 175 stores, 1,007 visits)
Entrance (p = .162)a
Not trained	76.8%	83.6%	85.8%
Trained	66.4%	83.1%	80.5%
Entrance or floor (p = .411)a
Not trained	90.5%	94.0%	94.3%
Trained	83.5%	93.3%	91.1%
Overall (entrance, floor, or counter) (p = .525)a
Not trained	94.3%	95.8%	97.7%
Trained	90.8%	96.3%	95.9%
Training uptake—Any employee trained (intervention group only n = 100 stores, 581 visits)
Entrance (p = .020)a
Not trained	82.6%	83.1%	84.5%
Trained	65.9%	82.5%	79.9%
Entrance or floor (p = .243)a
Not trained	90.8%	90.5%	92.0%
Trained	84.1%	93.9%	91.6%
Overall (entrance, floor, or counter) (p = .393)a
Not trained	93.8%	93.7%	98.4%
Trained	91.2%	96.8%	96.0%
Training uptake—Percentage of employees trained in store (entire sample n = 175 stores, 1,007 visits)
Entrance (p = .033)a
0%	61.3%	67.7%	70.2%
20%	58.3%	68.2%	68.9%
40%	55.4%	68.7%	67.6%
60%	52.4%	69.2%	66.4%
80%	49.4%	69.6%	65.1%
100%	46.4%	70.1%	63.8%
Parameter estimate	0.21	0.09	0.00
Entrance or floor (p = .274)a
0%	84.9%	88.4%	88.9%
20%	83.3%	88.7%	88.2%
40%	81.7%	89.1%	87.5%
60%	80.1%	89.4%	86.8%
80%	78.4%	89.8%	86.1%
100%	76.8%	90.2%	85.4%
Parameter estimate	-0.05	0.05	0.00
Overall (entrance, floor, or counter) (p = .434)a
0%	92.4%	93.9%	95.9%
20%	91.5%	94.2%	95.5%
40%	90.6%	94.5%	95.0%
60%	89.7%	94.8%	94.6%
80%	88.8%	95.1%	94.2%
100%	87.9%	95.3%	93.7%
Parameter estimate	-0.02	0.04	0.00
Training uptake—Percentage of employees trained in store (intervention group only n = 100 stores, 581 visits)
Entrance (p = .004)a
0%	57.8%	58.4%	60.4%
20%	53.3%	58.9%	59.2%
40%	49.1%	59.4%	58.0%
60%	44.8%	59.9%	56.9%
80%	40.5%	60.4%	55.7%
100%	36.2%	60.91%	54.5%
Parameter estimate	-0.16	0.08	0.00
Entrance or floor (p = .192)a
0%	81.8%	82.0%	83.7%
20%	80.2%	83.2%	83.6%
40%	78.8%	84.4%	83.5%
60%	77.3%	85.6%	83.3%
80%	75.8%	86.8%	83.2%
100%	74.3%	88.0%	83.1%
Parameter estimate	-0.07	0.08	0.00
Overall (entrance, floor, or counter) (p = .390)a
0%	90.4%	90.7%	94.9%
20%	89.7%	91.5%	94.3%
40%	89.0%	92.2%	93.8%
60%	88.3%	93.0%	93.3%
80%	87.6%	93.7%	92.8%
100%	86.9%	94.5%	92.2%
Parameter estimate	-0.01	0.06	0.00

Notes: Percentages can exceed 100% because of adjustment for covariates.

^a Two-way interaction among training uptake and assessment round, adjusted for state (Washington State vs. Colorado), type of store (recreational cannabis sales only vs. recreational and medical sales) and region (metropolitan area vs. outlying areas); two tailed p value.

The effect of training uptake on the refusal of pseudounderage patron buyers at the entrance to stores was most evident in Washington State. In the entire sample, the threeway interaction among training uptake, assessment round, and state was statistically significant, with store type and region being controlled for (Table 2). In Washington State, stores that used Train To Tend had a lower refusal rate at the store entrance at baseline than did stores that did not use the training, but, at both follow-ups, the stores using Train To Tend had much higher refusal rates, approaching the higher rates in the untrained stores. Training uptake, however, did not appear to affect refusal at entrance in Colorado stores. This three-way interaction was evident in the intervention group only, where only stores that did not use Train To Tend were classified as “not trained.” The training effect also emerged for the percentage of employees trained in refusals at entrance or at both entrance and on the floor combined (Table 2).

Table 2.

Adjusted estimated refusal rates by training uptake at various locations in recreational marijuana stores stratified by state

	Washington State			colorado
Variable	Baseline	3-month follow-up	9-month follow-up	Baseline	3-month follow-up	9-month follow-up
Training uptake – Any employee trained (entire sample n = 175 stores, 1,007 visits)
Entrance (p = .033)a
Not trained	57.9%	68.5%	72.3%	95.7%	98.6%	99.2%
Trained	40.3%	65.1%	60.4%	95.2%	101.0%	101.4%
Entrance or floor (p = .414)a
Not trained	84.4%	89.4%	88.9%	96.6%	98.6%	99.6%
Trained	71.0%	86.3%	82.4%	97.9%	100.4%	100.7%
Overall (entrance, floor, or counter) (p = .512)a
Not trained	89.5%	93.2%	96.1%	99.1%	98.4%	99.3%
Trained	84.8%	92.8%	92.1%	97.5%	99.7%	99.8%
Training uptake – Any employee trained (intervention group only n = 100 stores, 581 visits)
Entrance (p = .036)a
Not trained	67.2%	64.6%	67.7%	98.9%	102.0%	102.1%
Trained	37.6%	62.5%	57.7%	96.8%	102.4%	102.9%
Entrance or floor (p = .565)a
Not trained	84.3%	80.3%	83.7%	97.8%	100.5%	100.5%
Trained	70.8%	86.1%	81.9%	98.7%	101.1%	101.5%
Overall (entrance, floor, or counter) (p = .480)a
Not trained	87.5%	87.4%	97.9%	99.7%	99.7%	99.7%
Trained	84.9%	93.1%	91.6%	98.0%	100.0%	100.2%
Training uptake – Percentage of employees trained in store (entire sample n = 175 stores, 1,007 visits)
Entrance (p < .001)a
0%	62.7%	68.8%	71.4%	94.9%	101.1%	103.6%
20%	55.1%	69.4%	68.7%	95.6%	101.4%	103.5%
40%	47.6%	69.9%	65.9%	96.2%	101.8%	103.4%
60%	40.1%	70.5%	63.2%	96.9%	102.1%	103.3%
80%	32.5%	71.0%	60.5%	97.6%	102.5%	103.2%
100%	25.0%	71.6%	57.8%	98.2%	102.8%	103.1%
Parameter estimate	-0.24	0.16	0.00	0.17	0.15	0.13
Entrance or floor (p = .039)a
0%	85.8%	89.1%	89.7%	96.9%	100.2%	100.8%
20%	81.2%	89.5%	88.0%	97.7%	100.5%	100.9%
40%	76.7%	90.0%	86.3%	98.4%	100.7%	101.1%
60%	72.1%	90.4%	84.6%	99.2%	101.0%	101.3%
80%	67.5%	90.9%	82.9%	100.0%	101.3%	101.5%
100%	62.9%	91.3%	81.2%	100.8%	101.5%	101.7%
Parameter estimate	-0.14	0.11	0.00	0.12	0.10	0.09
Overall (entrance, floor, or counter) (p = .283)a
0%	92.9%	94.3%	96.3%	98.9%	100.3%	102.3%
20%	90.4%	94.6%	95.5%	99.2%	100.5%	102.1%
40%	87.9%	94.9%	94.8%	99.6%	100.8%	101.9%
60%	85.4%	95.2%	94.0%	99.9%	101.0%	101.8%
80%	83.0%	95.6%	93.3%	100.3%	101.2%	101.6%
100%	80.5%	95.9%	92.5%	100.7%	101.5%	101.4%
Parameter estimate	-0.09	0.05	0.00	0.06	0.05	0.03
Training uptake – Percentage of employees trained in store (intervention group only n = 100 stores, 581 visits)
Entrance (p < .001)a
0%	60.9%	61.0%	63.2%	97.0%	97.1%	99.3%
20%	51.8%	61.5%	60.5%	96.3%	97.4%	99.2%
40%	42.8%	62.1%	57.8%	95.7%	97.8%	99.1%
60%	33.7%	62.7%	55.2%	95.0%	98.1%	99.1%
80%	24.7%	63.3%	52.5%	94.3%	98.5%	99.0%
100%	15.6%	63.9%	49.8%	93.6%	98.8%	98.9%
Parameter estimate	-0.32	0.16	0.00	0.10	0.15	0.13
Entrance or floor (p = .112)a
0%	82.7%	82.5%	84.5%	99.9%	99.7%	101.7%
20%	78.9%	84.6%	84.0%	100.1%	100.2%	101.8%
40%	75.2%	86.7%	83.5%	100.4%	100.7%	101.9%
60%	71.4%	88.7%	83.0%	100.6%	101.2%	102.0%
80%	67.7%	90.8%	82.5%	100.8%	101.7%	102.2%
100%	64.0%	92.8%	82.0%	101.1%	102.1%	102.3%
Parameter estimate	-0.16	0.13	0.00	0.04	0.05	0.03
Overall (entrance, floor, or counter) (p = .460)a
0%	90.9%	91.0%	95.3%	99.4%	99.5%	103.8%
20%	89.0%	92.2%	94.6%	99.7%	99.9%	103.3%
40%	87.0%	93.4%	94.0%	99.9%	100.4%	102.9%
60%	85.1%	94.5%	93.4%	100.2%	100.8%	102.4%
80%	83.2%	95.7%	92.8%	100.5%	101.2%	101.9%
100%	81.3%	96.8%	92.1%	100.8%	101.6%	101.5%
Parameter estimate	-0.06	0.09	0.00	0.05	0.05	0.01

Notes: Percentages can exceed 100% because of adjustment for covariates.

^a Three-way interaction among training uptake, assessment round, and state, adjusted for type of store (recreational cannabis sales only vs. recreational and medical sales) and region (metropolitan area vs. outlying areas); two tailed p value.

The increase in refusal rate occurred between baseline and the 3-month follow-up, when stores were invited to use the training. There may have been some decline in refusal rates from the 3- to the 9-month follow-up in the trained stores, perhaps indicating that some training effect lacked persistence.

Discussion

The RMV training may have improved compliance with state ID requirements at recreational marijuana stores that used it, especially in Washington State. The failure to find an overall improvement, and instead to observe a training uptake effect, may have stemmed from two circumstances. First, ID checking was nearly universal, and the large majority of pseudo-underage patron buyers were refused somewhere in the stores. This might be motivated by state regulators working with stores and industry concerns about state compliance checks (Wiens et al., 2018). The nearly universal ID checking could have created a strong ceiling effect that limited the ability for RMV training to improve sales practices. However, stores in Washington State were less likely to refuse pseudo-underage patron buyers at the store entrance, and it was in that location that RMV training appeared to improve refusal rate. The tendency to check IDs less frequently in Washington State stores at baseline may stem from it being illegal for minors themselves to enter the store, not just for store personnel to permit or refuse entry. RMV training may have convinced some Washington State store owners or managers that it would be better to check IDs at entry and not after customers reach the sales floor, but as this study was not designed to measure state-level policies, the reason for differences is uncertain.

Second, only 42% of stores in the intervention group used the RMV training, and stores that used it appeared to practice less responsible sales behaviors at baseline (i.e., had lower baseline refusal rates) than did stores not using it. Managers at poorly performing stores may have believed that employees would benefit from RMV training and thus promoted its use. Once employees were trained, these stores improved their refusal rates to levels similar to the initially high refusal rates at the untrained stores. The benefit of training seemed to occur when more employees in a store completed it, which increased the likelihood that pseudounderage patron buyers encountered a trained employee. This is an argument for requiring RMV training of all recreational marijuana store personnel to achieve maximum benefit. Currently, Alaska, Illinois, Oregon, and Massachusetts require RMV training in some form for all store employees. Colorado incentivizes it for recreational marijuana stores and requires it for stores providing home delivery of medical marijuana (and for hospitality licensees where cannabis can be consumed on premises).

There was some suggestion that the impact of training uptake did not fully persist at the 9-month follow-up. Similar declines were seen in studies of some RBS training (Toomey et al., 2008, 2017), but the authors observed a persisting increase in refusal rates from an RBS training for servers (Woodall et al., 2018). Such declines, when they occur, may stem from a multitude of factors, including decay in knowledge and skills, decrease in motivation, lack of management support, incentives to sell, and absence of professional norms for responsible sales practices. Stores might set house policies, make managers and employees aware of them, communicate about responsible sales practices, and create a sense of professionalism to ensure persisting benefit of RMV training (Toomey et al., 2017).

It is possible that the RMV training was not completely responsible for improvements in the stores that used it. Managers of under-performing stores could have made several changes that, along with training, improved refusal rates. One change could have been to check all customers’ IDs first at the door and then again at the counter. This double ID checking procedure, sometimes used in nightclubs and frequently performed by bouncers whose primary job is checking IDs, was more common in Colorado than in Washington State stores. Increasing refusals at the entrance might prevent buyers from reaching the counter where staff could feel pressure to sell marijuana. Managers may have expressed their strong support for responsible sales practices, including checking IDs, which has influenced training success in alcohol outlets (Ker & Chinnock, 2008; Shults et al., 2001). However, testing within the RMV training showed that employees improved their ID checking skills (Buller et al., 2019b), so RMV training was probably responsible for some of the improvement.

There were potential limitations to the trial. As noted, not all stores in the intervention group used the training, and not all employees in a store completed it. Thus, we could not be certain that the pseudo-underage buyer interacted with trained store personnel. This may generalize to normal policy circumstances. When states incentivize training, some but not all stores use the training, and managers may decide to train some but not all personnel. When training is required, sales personnel are usually given a grace period to complete it (e.g., 30–60 days), and some stores may flaunt the law and not train employees. The pseudo-underage patron protocol had buyers not show an ID, which may be a very noticeable violation of ID requirements. However, a protocol where underage buyers showed their actual IDs did not yield different refusal rates (Buller et al., 2019a). Small differences in state laws may have led to different approaches to ID checking that affected training outcomes, but both states make it illegal for underage persons to be in the store and require age verification. The pseudo-underage patron protocol was an observational measure that presented sales personnel with an overt circumstance where they should refuse sales, making it superior to self-reports.

The trial was performed in the first two U.S. states that legalized recreational marijuana. Although this may limit generalizability, subsequent states promulgated ID regulations similar to those of Colorado and Washington State. The recreational marijuana regulations and markets will undoubtedly evolve (e.g., four states allow home delivery; Leafbuyer Writing Team, 2020). Despite changes in the alcohol market (Ecklund et al., 2017), RBS training has improved responsible sales practices and reduced intoxication and alcohol-impaired crashes in some studies (Dresser, 2000; Graham, 2000; Scherer et al., 2015), particularly when the training is delivered online (Woodall et al., 2018) and with strong management support (Ecklund et al., 2017; Shults, et al., 2001).

An advantage of online training is that it can be easily updated for market changes (e.g., content was added to Train To Tend for home delivery and onsite consumption to maintain its applicability). The findings pertain to online training with interactive multimedia learning activities, which we believe creates a high-quality, high-fidelity, engaging learning environment (Buller et al., 2019b).

These findings may not apply to RMV training in other forms, such as in-person or online didactic presentations. Also, our training was largely focused on responsible practices of sales personnel, not management practices (e.g., setting store policies). Management support can improve responsible alcohol sales practices (Ker & Chinnock, 2008; Shults et al., 2001). Offsetting these limitations were a number of strengths, including the large, random sample of state-licensed stores from diverse regions, prospective randomization to treatment following baseline, blinding of pseudo-underage patron teams, and multiple assessments of stores.

From older markets, such as the alcohol market, there may be lessons from interventions that can mitigate potential harms in the new recreational marijuana market. Training of store personnel in responsible sales practice is one such intervention (Gripenberg Abdon et al., 2011; Gripenberg et al., 2007; Ker & Chinnock, 2008; Wallin et al., 2005). Online training, in particular, may make training effects predictable and be acceptable and preferable to in-store training of store personnel (Buller et al., 2019b; Ker & Chinnock, 2008; Woodall et al., 2018). However, such training may be effective only if stores use it and if employees are motivated to implement responsible sales practices.