Multidimensional Temptation: Field Evidence

Respondents in the study complete a survey in which they submit preferences over seven meal platforms (which differ in the number of meals offered). After completing the survey, all respondents will be invited to a second stage. In the second stage, participants receive a link to one of the seven food ordering platforms. The specific food ordering platforms are partly determined by the survey responses (e.g., orderings that participants submitted) and partly determined by chance. We will further track the number of meals respondents collect in colleges.

Study participants are students in 6 colleges at the University of Oxford who will be invited for a meal challenge: limiting their consumption of animal products and/or refined carbs for 5 days in a row, focusing instead on plant-based options. They will be offered 5 free lunch meals at their dining hall for a week and asked to order their meals on a food ordering platform designed specifically for this project. The meals may come from one of three meal categories: Daily Harvest (no meat/fish, limited carbs), Carb Powerhouse (no meat/fish, rich in carbs) and Carnivore Corner (meat/fish, limited carbs). We study participants' preference ordering over 7 different versions of the platform, which vary in the number of meal categories displayed (i.e., the menu). This design allows us to study (i) the trade-offs between commitment and flexibility (i.e., restricted vs. larger menus); (ii) the nature of commitment choices. We collect data on participants' motives and beliefs to interpret these measures. Participants are then randomly assigned one version of the platform based partly on their decisions and partly on chance (50% will receive the unrestricted version of the platform for sure and 50% will receive a platform based on how they ranked the 7 different versions). Subsequently, we study the meal orders made by participants on the assigned platform, their likelihood of picking up their meals, and their subjective experiences and attitudes after the challenge period is over.

2025-02-09

2025-02-14

(i) Indicator=1 if the participant put the unrestricted platform (showing all 3 meal categories) at the top of their ranking (alone) and =0 otherwise;
(ii) Indicator=1 if the participant put the restricted platform with Daily Harvest only at the top of their ranking (alone) and =0 otherwise;
(iii) Indicator=1 if the participant actually chose Daily Harvest when facing the unrestricted platform and =0 otherwise;
(iv) Number of meals collected (between 0 and 5).

Other outcomes will be constructed based on the survey responses tailored to separate stochastic from cumulative temptation. We will calculate a global temptation index that measures the number of times a respondent strictly prefers to eliminate the singleton meal category from non-singleton menus that include it; (ii) calculate a strictly global temptation index, which imposes additional conditions on the preferences over the meal categories; (iii) define categories of menu preferences (standard, flexibility, no self-control, self-control, global commitment). Please consult the attached preregistration document for more details.

(i) Indicator=1 if the participant submitted an order on the platform and =0 otherwise;
(ii) Indicator=1 if the participant collected all 5 meals and =0 otherwise;
(iii) Indicator=1 if the participant collected at least one meal and =0 otherwise.

Please consult the attached preregistration document for more details.

We aim to recruit 400 students across 6 colleges, targeting roughly 12% of the student body within each college. The study consists of 4 main parts.

(i) Main survey (20-25 mins): Collection of socio-demographics background variables, meal habits, and introduction to the meal challenge and food ordering platform. Elicitation of preferences for the menu shown on the platform via iterative ranking and monetary valuations. Administered the week of 03/02/2025.
(ii) Meal ordering: Order of 5 meals on the platform that was randomly assigned to the respondents. The orders need to be submitted in the morning of 10/02/2025.
(iii) Meal pickup: Consumption on site of the meals ordered between 10/02/2025 and 14/02/2025 (each day, Monday to Friday during college hall lunchtime hours).
(iv) Follow-up survey (10-15 mins): Feedback survey on experiences. Sent at the end of the challenge (from 14/02/2025).

There are two sources of randomization:
(1) Randomization of menu assigned on the platform: This randomization will be performed in R after the main survey took place. We will perform the randomization at the college level.
(2) Information treatment: In the main survey, we will randomize a message at the college level. The control message will simply remind participants of the two dimensions of performance in the challenge (choice of plant-based options and hall attendance). The treatment message will say that an AI model expressed uncertainty about their performance in the challenge (accounting for both dimensions). We will assess the impact of the message on (i) their self-reported motivation (primary); (ii) their menu preferences (primary); (iii) their expected meal category choice (secondary); (iv) their expected number of meal pickups (secondary); (v) their chosen meal category if they got the unrestricted platform (primary); (vi) their actual number of meal pickups (primary); (vii) whether they picked up all 5 meals (secondary).

Randomization was performed via survey software (Qualtrics) and statistical software by computer (Python and R).

Individual

No

Target: 400 participants (selected partly for logistical reasons)

Target: 400 participants (selected partly for logistical reasons)

Target: 200 participants per treatment arm (selected partly for logistical reasons)

With 400 subjects (200 in each treatment arm), we will have 80% power to detect a 13 percentage point difference or higher in the probability of collecting all 5 meals, assuming a reference value of 0.6 in the control group (assumed based on data from a prior experiment), based on a two-sided test of proportions with alpha = 0.05.