The primary aim of the study was to assess the effect of the presence of dynamic descriptive social norm messages in the workplace cafeteria on the likelihood of cafeteria customers selecting a meat-free meal option. We hypothesized that the proportion of weekly plant-based meal sales would be higher in the intervention sites compared to the control sites.

We also hypothesized that the intervention would be more effective in cafeterias that served offices, had high plant-based meal availability, and high intervention fidelity. Our secondary aim was to assess whether the intervention had any unintended consequences, such as a decrease in revenue, that could decrease its acceptability by food retailers. We hypothesized that the intervention would have no impact on overall sales. Finally, we aimed to assess the salience and perceived credibility of the dynamic descriptive norm messages through opportunistic interviews with cafeteria customers. As these interviews were exploratory, we did not have a pre-specified hypothesis for their outcome.

We conducted a two-arm, parallel, randomized controlled trial in worksite cafeterias with the assistance of the central management team of a large global catering company that provides contract catering services to various companies and institutions across the United Kingdom. The trial ran in 27 cafeterias for 8 weeks, starting on 10 October 2022 and ending on 03 December 2022.

The inclusion criteria required each cafeteria to (a) be based in the UK, (b) have electronic point-of-sale tills operated by the catering company, (c) be able to provide data at a detailed enough level to identify specific meals sold, and (d) offer hot main meal options (in line with the base menu for the catering company). The sample size was based on pragmatic factors, rather than a power calculation, including willingness of worksite cafeteria managers to join the trial, the size of the worksite they were serving, contracting company, and baseline meat-free meal sales. Cafeterias were initially invited to participate in the study via an email sent to them by the Head of Sustainability of the catering company and were put into contact with the research team once they expressed initial interest.

The cafeterias that took part in this study all had a daily two-hour lunch service during which hot main meals and sides, salad bar, sandwiches, wraps, beverages, and confectionery were available for purchase. Each cafeteria curated their selection of food based on their customer profiles by choosing from a standard list of recipes provided by the central management of the catering company. The hot meals differed daily, and customers could view a weekly menu in advance. The prices of the hot meals (both meat-based and meat-free) were similar to each other, with a maximum of 5–10 pence difference. All cafeterias were self-service, with customers walking through each station and choosing what they wanted to eat and paying for it at the tills before sitting down to consume the food. All cafeterias operated within a single company’s facility, only serving employees of that specific company.

After recruitment, eligible worksite cafeterias were randomized with an allocation ratio of 1:1 to one of the following conditions: control (no message) and intervention (social norm messages displayed prominently in the cafeteria). The lead author conducted the randomization via random number assignment using STATA version IC 16.1 [23], based on a random seed generated online (https://bit.ly/stata-random) to enable replication. The necessary code for the randomization sequence was written and run on a separate do-file, and the researchers did not view the numbers and conditions assigned to the cafeterias until the complete code was run. There was no stratification or limitation set for the process.

It was not possible to blind researchers or worksite cafeteria managers to the intervention allocation due to the nature of its implementation. However, cafeteria users were not made aware of the research during the trial period, and no explicit individual consent of cafeteria customers was obtained since no personally identifiable information was collected, and the data analyzed was at worksite-level. The catering company obtained verbal consent from the contracting companies where the worksite cafeterias were housed and from the cafeteria managers that expressed interest in joining the trial. Ethical approval was given by the University of Oxford Central University Research Ethics Committee (Reference: R72710/RE001).

The intervention consisted of three main components (see Fig. 1). The first component, “More and more of your [company name] colleagues are choosing veggie options,” conveyed the dynamic descriptive norm by implying an increased adoption of the target behavior (choice of vegetarian meal option). In line with evidence that higher context specificity increases the potential influence of social norms [24], the message referred to a specific and relevant peer groups, i.e., other co-workers of the contracting company in which the worksite cafeteria is located with the phrase “your [company name] colleagues.” The second component, “Join them today” aimed to reinforce the norm message with a call to action that invites the cafeteria customers to choose a plant-based meal “today,” grounding the performance of the normative behavior to a specific time [25, 26]. The final component, “Spotted the star? Look for the star on our most loved veggie options” added an element of gamification that urged customers to identify an easily perceivable feature (the star). It also highlighted the vegetarian dishes as the “most loved” option, which built on previous evidence that suggested drawing attention to plant-based dishes on menus by marking them as “dish of the day” or “chef’s recommendation” may increase their likelihood of getting selected [27].

Intervention materials used in the study. a Large free-standing banner promoting the intervention message; b floor stickers placed in the dining area; c A4-sized posters with a star sticker indicating the vegetarian option on the menu. Company branding has been redacted for anonymity

The materials and visual aspects of the intervention were developed in collaboration with the marketing team and central management of the catering company. The materials consisted of a large free-standing three-sided banner placed at the entrance of the cafeteria, floor stickers that guided the customers from the entrance to the hot main meal buffet, A4-sized posters that were placed next to or above the buffet, and star stickers that were placed next to the vegetarian options on the menu. The banners, floor stickers, and posters all featured the three message components specified below and were all designed to make the messages as salient as possible by increasing exposure and catching attention (see Fig. 1).

All worksite cafeteria managers that provided verbal consent were invited to an online conference call where the research team outlined the trial design, purpose, and intervention materials. Following randomization, managers whose cafeterias were allocated to the intervention condition were invited to a second online call where researchers provided detailed information on how to implement the intervention and the procedure for fidelity checks.

Intervention cafeterias received their materials (i.e., banner, floor stickers, posters, menu star stickers) one week prior to the start of the trial and guidance on how to place the materials (banners at the entrance, floor stickers leading to hot buffet, and posters at the hot buffet) and marked the vegetarian options on their printout menus every day with the star stickers.

All cafeterias, including those in the control condition, received four phone calls over eight weeks from researchers. All cafeterias were asked to report any site closures, till malfunctions, promotions and special events, unexpected changes to the preplanned menus, ingredient and supply shortages, or other disruptions and changes to the purchasing environment that could have affected footfall and sales. Intervention cafeterias were additionally asked whether the materials were still intact and on display, whether vegetarian options were being labelled daily, whether the catering staff had any feedback on the intervention, and whether any customers asked questions about the intervention materials. Intervention cafeteria staff were also required to send weekly photographs of the materials and menus to check that the intervention continued to be implemented correctly. One cafeteria manager was uncontactable, and one cafeteria had a new manager come into the role who did not receive information about the intervention during the handover. These two cafeterias were both part of the intervention condition and were treated as lost to follow up and excluded from analysis (Fig. 2).

CONSORT flowchart illustrating the progression of cafeterias through each stage of the trial, including enrollment, allocation, follow-up, and analysis

During the final 2 weeks of the intervention, the research team visited 10 out of the 12 intervention cafeterias to observe the implementation in person and receive feedback from catering managers and staff. Team members arrived at each cafeteria shortly before lunch service and observed the customers’ arrival to the cafeteria, their food purchasing patterns, and their interactions during consumption. Opportunistic semi-structured interviews with willing cafeteria customers were conducted, who were asked whether they noticed the intervention materials, whether they could recall the content of the messages, whether they found this content to be credible and believable, whether they had a positive or negative reaction to the messages, and whether they made changes in their consumption pattern as a result. Team members recorded answers in writing using questionnaire templates (see Additional File 1: S1).

The catering company provided the research team with sales data that were recorded via electronic point-of-sale tills throughout the trial. Hot main meals, wraps, salads, soups, sandwiches, savory snacks, starters, and jacket potatoes were coded as either meat-based or meat-free depending on their ingredients using a series of keywords (e.g., “chicken,” “fish,” “lamb” or “Quorn,” “plant-based,” vegan”). Food items that do not constitute a meal such as sides, confectionery, desserts, beverages, and condiments were excluded from the data. All products were individually checked by a member of the research team for potential errors in coding (e.g., coding a meat-including item as meat-free). Clarifications were asked from catering managers if the contents of a dish were not clear from its name. Following coding, the total number of meals sold every day and every week at each site were calculated, for both the 8-week period that preceded the trial which constituted the baseline, and the eight-week period of the trial itself. The total number of meals was used as a proxy to understand whether there were changes in footfall and overall meal selection in cafeterias during the intervention period. The proportion of meat-free dish sales was calculated for each day and each week for all participating worksite cafeterias for baseline and trial periods, which was the main outcome of interest.

The catering company provided the research team with sales data for the 8-week period that preceded the intervention. Baseline meal sales were entered into all regression models as a control variable.

This was used as a proxy to estimate the socioeconomic status of the cafeteria customers, with offices assessed as having a larger concentration of higher SES customers and factories/manufacturing sites as having a larger concentration of lower SES customers. Worksites were classified as an office site if their main activities included financial, accounting, and customer services and telecommunications, and as a manufacturing site if they focused on production, processing, distribution, and storage of various goods.

Each site was coded for its intervention fidelity based on its adherence to the intervention period based on data gathered from phone calls, photographs, and cafeteria visits. Cafeterias that displayed the materials correctly throughout, answered all phone calls, and provided weekly photos were coded to have high intervention fidelity, whereas those that missed one or more calls or did not send photos each week were coded to have low fidelity, creating a dummy variable.

Any closures or unexpectedly low footfall days due to bank holidays, worker strikes, or worksite-specific events were recorded, and observations from those days were excluded from the data.

Any issues with recording sales using the electronic point of sale devices (e.g., due to internet connection failures or malfunctioning devices) were recorded for each cafeteria, and observations from those days were excluded from the data.

The statistical analysis plan specifying hypotheses and primary and secondary outcomes was pre-registered on OSF before any data cleaning or analysis was conducted (https://osf.io/6wfgu). The analysis was done using STATA version IC 16.1 [23]. The data received from the catering company were cleaned, and discrepancies in records from each cafeteria (e.g., different ways of entering sale dates) were resolved prior to any analysis. Non-meal food items (e.g., confectionery—as explained in the Measures section above) were excluded from the data. Observations from weekends and national holidays were excluded from the data as they only occurred in a small number of sites and had a very low number of observations. Days where cafeterias experienced a till malfunction, closure, or strikes were excluded from the data, and a variable that indicated that the cafeteria had a shorter week was created. Overall, 7 intervention and 5 control cafeterias experienced at least one day of till malfunction, closure, or strikes, resulting in a total of 47 baseline period days/cafeteria (out of 1425 days/cafeteria) and 58 days/cafeteria (out of 1375 days/cafeteria) being excluded from the final analysis.

The primary research question was whether the presence of dynamic descriptive social norm messages in the workplace cafeteria changed the likelihood of cafeteria customers selecting a meat-free meal option, when controlling for baseline percentage of vegetarian main meal sales and length of operational week. Following from this, we also explored whether the effect of the presence of dynamic descriptive social norm messages in the workplace cafeteria varied when interacted with baseline sales, worksite type, and intervention fidelity. To answer these questions, multilevel mixed-effects linear regressions with restricted maximum likelihood and with Kenward-Roger method for degrees of freedom were run with autoregressive (AR) structure of order for within-group errors where the weeks were used as an integer-valued time variable to order the observations within groups and to determine the lags between successive observations. The p-value thresholds for significance were set at 0.05 for the primary and at 0.01 with a Bonferroni correction for multiple testing for the secondary research question. The weekly percentage of meat-free sales during the intervention period was first regressed on trial arm (control (N = 13) versus intervention (N = 12)), entering baseline percentage of meat-free meal sales and length of operational week as covariates. Then, another regression model was created adding type of worksite (office versus manufacturing) and baseline percentage of meat-free meal sales as interaction terms, with random effects varying according to each cafeteria. The regressions were repeated with a per-protocol analysis excluding 2 cafeterias from the intervention arm that had low intervention fidelity for a sensitivity analysis.

The secondary research question examined whether the overall sales of meals by the cafeterias changed with the presence of dynamic descriptive social norm messages. For this question, the above regression model was used again, but the dependent variable was replaced with weekly total meal sales during the intervention period, and the baseline variable was calculated based on total meal sales. The regression was repeated with a per-protocol analysis excluding 2 cafeterias from the intervention arm that had low intervention fidelity for a sensitivity analysis.

We also explored how cafeteria customers perceived the dynamic descriptive social norm messages in the workplace cafeteria. For this question, opportunistic interviews with cafeteria customers were descriptively analyzed for the frequency of recalls of the intervention material and message, the number of customers who thought the message was believable and credible, whether the message was able to motivate individuals to change their behavior, and the priorities of individuals when choosing a meal at their worksite cafeteria. These interviews took place during the last 2 weeks of the intervention trial period during regular lunch service hours when customers were most likely to visit the cafeterias.

The current analysis deviated from the original statistical analysis plan in several aspects. We initially intended to limit our analysis to hot main meal sales in each cafeteria, but due to the low sales of these items and the comparatively high sales of wraps, sandwiches, salads, and pasties in each cafeteria, the scope of the sales data which we analyzed was extended. The original plan also aimed to assess the availability of meat-free hot main meals in each cafeteria and enter this as an interaction variable to predict the effect of the intervention on meat-free sales. However, both due to the inclusion of other food items in the analyzed data and the lack of consistent information about the content of menus offered at each cafeteria, we decided that it was not possible to meaningfully measure the availability of meat-free items. Finally, the original plan intended to assess the change in the environmental impact of meals sold as a result of the intervention. However, since there was no evidence of a main effect of the intervention on meal sales, we did not go ahead with this analysis.