The Role of Motivated Beliefs in the Carbon Footprint of Food Consumption

We consider three experimental conditions, differing in the public policy implemented at the final meal-box choice: no intervention (control), a carbon tax, or a label and boost intervention. All conditions share the following common structure:
i) Subjects assemble a meal box containing four meals for two people.
ii) We elicit subjects' beliefs about whether the carbon footprint of their chosen meal box is lower than the average carbon footprint of the meal boxes selected by participants in the baseline session.
iii) Subjects receive a first signal regarding their footprint relative to the reference. The signal's accuracy is randomly determined (i.e., it is accurate or inaccurate with a given probability), and we measure how subjects update their beliefs.
iv) Subjects receive a second signal, whose accuracy is again randomly and independently determined, and we measure how subjects update their beliefs.
v) Subjects have the option to revise their selected meal box.
vi) We re-elicit subjects' beliefs about the carbon footprint of their (possibly revised) meal box.
vii) The relevant public policy is implemented, or not (control condition), and subjects again have the option to revise their meal box.

The three conditions differ only at step vii):
- Control: no intervention is implemented at the final choice stage.
- Carbon tax: a tax of €300 per ton of CO2 (the value recommended in the 2025 Quinet report) is applied to the price of each available meal, proportionally to its carbon footprint. The set of available meals is unchanged from the earlier choice tasks and identical to that of the control condition.
- Label and boost: meals display the Carbon-Score label (a letter grade from A to E, color-coded from green to red), provided by Quitoque, the company supplying the meal boxes, and derived from a transformation of each meal's carbon score. The label is accompanied by a boost message indicating that one effective way to reduce the climate impact of food choices is to favor white meat over red meat and to choose meals richer in plant-based protein (soy, peas, lentils) with less meat. This recommendation is consistent with guidance from ADEME and with the scientific consensus reflected in the EAT-Lancet Commission report.

2026-07-10

2026-10-31

Our primary analysis will focus on two outcomes: (i) subjects' beliefs about whether the carbon footprint of their chosen meal plate is lower than the average footprint of meal boxes selected by participants in the baseline sessions, and how these beliefs are updated following the signals subjects receive; and (ii) the carbon footprint of the meal boxes selected by participants at each choice stage.

Beliefs are elicited at stage ii), before any signal is received, providing a baseline measure, and again at stages iii) and iv), immediately after each signal. Comparing beliefs across these stages allows us to assess whether subjects update their beliefs in the direction of the signal received, consistent with Bayesian updating, or in an asymmetric or insufficient manner, consistent with motivated reasoning. Beliefs are elicited once more at stage vi), after subjects have had the opportunity to revise their meal box, to assess whether belief updating persists once subjects have acted on the information received.

Meal boxes are chosen at stages i), v) and vii), and we observe the associated carbon footprint at each stage. Comparing meal-box carbon footprint at stage i) (before any signal) with stage v) (after both signals) allows us to assess whether receiving information about one's carbon footprint leads subjects to revise their food choices. Comparing meal-box carbon footprint at stage v) with stage vii) (after the policy is implemented, or not) allows us to assess the effectiveness of the carbon tax and of the label-and-boost intervention. In addition, we will compare the change in footprint between stages v) and vii) across the three experimental conditions (control, carbon tax, label-and-boost) to assess the relative effectiveness of the two policies, and whether this effectiveness depends on how subjects previously updated their beliefs.

In addition to the primary outcomes, we will use a post-experimental questionnaire to construct the following secondary outcomes: (i) subjects' self-reported attitudes toward health and the environment; (ii) subjects' sociodemographic characteristics; and (iii) subjects' responses to control questions assessing their understanding of, and behavior during, the experiment.

The questionnaire collects subjects' attitudes on health and the environment (Kantar’s lifestyle questionnaire). We will use these self-reported attitudes to examine whether the belief revision and behavioral change documented in the primary analysis are associated with subjects' environmental and health attitudes.

Sociodemographic characteristics collected in the questionnaire (e.g., age, gender, education, household composition, income) will be used to examine heterogeneity in belief revision and behavioral responses across subgroups.

Finally, the control questions included in the questionnaire assess participants' understanding of the experimental procedures, in particular the belief-elicitation mechanism, as well as their self-reported behavior during the task. These responses will be used to verify that the documented patterns of belief revision are not driven by a lack of understanding of the procedure, and to test whether the degree of comprehension moderates the extent of belief updating.

Baseline sessions
A few days before the start of the experiment, around 50 participants are invited to the laboratory (GAEL laboratory) to assemble a meal delivery box containing four meals for two adults, with a value of €60, from the range offered by the company Quitoque (approximately 30 recipes suitable for all types of diets). The average carbon footprint of the boxes selected during the baseline session is then calculated. This value serves as the reference benchmark for the remainder of the experiment.

Experiment
Participants are invited for an experimental session divided into three phases. The objective of the first phase is to analyze how individuals revise their beliefs about the carbon footprint of their diet, distinguishing individuals who hold erroneous beliefs due to limited information from those whose belief revision reflects motivated reasoning. Beliefs are elicited using an incentive-compatible mechanism, explained to participants during a pre-experimental training session.

Pre-experimental training: participants estimate the probability that a meal box has a lower carbon footprint than the average carbon footprint of the boxes selected during the baseline session. This belief is elicited using a mechanism designed to incentivize honest reporting: the participant states their belief, this statement is compared to a randomly drawn number, and depending on the outcome of this comparison, the payment depends either on the actual realization of the event or on an independent random draw. This procedure ensures that the participant's expected payoff is maximized by truthfully reporting their belief. A non-incentivized practice trial is proposed to familiarize participants with the procedure, and participants cannot proceed to the main task until they have demonstrated understanding of the mechanism.

The first phase then proceeds as follows:
i) Participants assemble a meal delivery box containing four meals for two people, valued €60, from the Quitoque range (approximately 30 recipes suitable for all dietary needs). Participants are informed that, at the end of the experiment, a random draw will determine with 10% probability whether they receive one of the boxes they assemble. For these randomly selected subjects, any unspent amount from the €60 will be added to their final earnings.
ii) Participants estimate the probability that their box has a lower carbon footprint than the average carbon footprint of the boxes selected in the baseline session, using the elicitation mechanism described above.
iii) Participants then receive a noisy signal (framed as either a positive or a negative message) about the environmental impact of their chosen box. They are told that this signal is accurate with probability two-thirds. The noise introduced by this imperfect accuracy leaves room for motivated beliefs revision: it allows us to distinguish subjects who update their beliefs symmetrically, regardless of whether the signal is favorable or unfavorable (consistent with Bayesian updating), from those who discount unfavorable signals more than favorable ones (consistent with motivated reasoning). Participants report their belief again.
iv) A second noisy signal, also accurate with probability two-thirds and drawn independently of the first, is sent regarding the environmental performance of the same initial choice. Participants report their belief once more.

This sequence of signals and belief elicitations allows us to test whether belief revision follows Bayesian updating or instead reflects motivated reasoning.

The second phase focuses on behavioral revision:
v) After having revised, or not, their beliefs, participants may modify the contents of their delivery box.
vi) Participants then report their beliefs regarding the environmental performance of this (revised) choice.

The third phase examines the moderating role of motivated reasoning mechanisms on the effectiveness of food policies. Subjects are randomly assigned, between subjects, to one of three experimental conditions (one and only one condition per subject):
- A control group, in which no policy is implemented.
- A group for which a carbon tax of €300 per ton of CO2 (as recommended by the 2025 Quinet report) is applied to the price of each available meal, in proportion to its carbon footprint. The set of available meals is unchanged relative to the previous choice tasks and identical to that of the control group.
- A group exposed to a Carbon-Score label supplemented by a boost message. The Carbon-Score label (letter grades from A to E, color-coded from green to red) is provided by Quitoque and derived from a transformation of each meal's carbon score. The boost message states that one effective way to reduce the climate impact of food choices is to favor white meat over red meat and to choose meals richer in plant-based protein (soy, peas, lentils) with less meat (a recommendation consistent with ADEME guidance and the scientific consensus reflected in the EAT-Lancet Commission report).

vii) Following the policy implementation, participants are invited to revise the contents of their delivery box for the third time.

Questionnaire
In a post-experimental step, participants complete a questionnaire collecting their sociodemographic characteristics and their attitudes on several dimensions, particularly health and the environment. The questionnaire also includes control questions designed to assess participants' understanding of, and behavior during, the experiment.

With this experimental design, and based on the existing literature, we aim at testing the following hypotheses:

Hypotheses related to motivated reasoning:
H1: Individuals hold overly optimistic beliefs regarding the environmental performance of their own meal boxes;
H2: Individuals update their beliefs asymmetrically, assigning greater weight to favourable information than to unfavourable information regarding the environmental performance of their meal boxes;

Hypotheses related to behavior:
H3: Individuals exhibiting stronger motivated reasoning mechanisms make smaller downward revisions to the carbon footprint of their meal boxes following unfavourable information;

H4a: Informational intervention is less effective among individuals exhibiting stronger motivated belief mechanisms;

H4b: Motivated reasoning mechanisms moderate the effectiveness of informational interventions more strongly than the effectiveness of carbon taxation.

Indeed, taxes alter incentives directly and therefore leave less room for motivated reinterpretation than informational interventions.

Not available

Randomization made by computer

Individual

No

NA

At least 500 subjects in total (150 per experimental condition and 50 for the baseline sessions).

We will recruit a minimum of 150 subjects per experimental condition, and 50 subjects for the baseline sessions.

NA

Pre-analysis plan.pdf