This research data contains data from across 2017 and 2018 on vegetarian sales and model estimates from cafeterias at University of Cambridge colleges A and B. The aim of the study was to investigate the effect of meal presentation order on vegetarian sales. We investigated this by changing meal order between “VegFirst” – positioning the vegetarian option first – and “MeatFirst”, in two studies involving 105,143 meal selections.

The two studies reported here consisted of multiple treatment reversal design experiments, swapping the order each week (or month) in which customers were presented with vegetarian and meat main-meal options at lunch and dinner times. The experiments were conducted in two University of Cambridge (UK) college cafeterias; a college is similar to a hall of residence or dorm. College A is a graduate college with over 600 students. College B has over 900 students, both undergraduate and graduate. Both colleges admit students of any gender identity. Meals are not included in the tuition or accommodation fees: students can choose to eat in the college cafeteria, cook their own meals or eat at other establishments. Students pay for meals by swiping their university cards. The cafeterias are largely self-service: students take a tray, view the different meal options available, and ask the serving staff for their preferred meal and side dishes. Students serve themselves salads, desserts and other cold items. In College A, diners have to walk past all options to reach the cash register. However in College B, the entire cafeteria is square rather than rectangular (with an island in the middle with salad components) and diners do not need to walk past all the options to reach the cash registers on the left hand side of cafeteria. These studies were approved by the University of Cambridge Psychology Research Ethics Committee (PRE.2016.100). Consent was obtained from catering managers; diners were not informed about the studies.
The primary outcome variable was the number of vegetarian meals sold at each mealtime, expressed as a percentage of the total meal sales. Salads, sandwiches and side dishes were not included. College A provided four options at lunch and five at dinner; sometimes a second vegetarian or vegan option was provided but this did not count towards the sales of the focal vegetarian option (SI Table S1). College B had a third main option, placed towards the back of the cafeteria. In summer term 2017 this third option was always meat at lunch and dinner (SI Table S2), but starting in autumn term 2017 at lunchtimes a vegan option was provided (SI Table S3). Similarly, the vegan sales did not contribute to the vegetarian sales considered in our analysis (see SI Tables S10 and S11). Following the recommendation of Simmons et al,17 we discuss results from both univariate and multivariate models (Supplementary Methods). Sales data were downloaded from the online catering platform Uniware18. Many individuals buy more than one meal from their college cafeteria over a term. In the absence of individual-level data, each meal selection was treated as independent. While this approach has been used in numerous other studies6, it adds uncertainty to p-value estimates. We therefore focused primarily on the effect size of our intervention, presenting the odds ratios and 95% confidence intervals and McFadden’s pseudo R2. The odds ratio – i.e. the effect size – was calculated by taking the exponential of the model estimate. Model diagnostics were used to check that the models did not violate any regression assumptions. We carried out all analyses in R 3.519.