Additional data collection to: Quantifying the role of greenhouse gas emissions in consumption choice

2021-10-26

2021-11-05

The data collected in the second data collection wave can be pooled with data from the first
data collection wave. Differences in meal tastes between July and November are controlled for
due to individual by meal fixed effects included in each analysis. A total of eight observations
are made for each individual, as for each of the four meals, the change in WTP taking place both from (1) the first to the second WTP inquiry and (2) the first to the third WTP inquiry is
observed. This approach maximizes power, leading to the following number of observations for
each treatment effect (provided 300 participants can be recruited in the second data collection
wave):
1. Change in demand occurring in reaction to being made aware of emissions (by being asked
to guess emissions caused), without being shown emission labels: 1240 observations (155
participants in Group C observed twice in the first data collection wave)
2. Change in demand occurring in reaction to being made aware of emissions and then shown
emission labels: 1240 observations (155 participants in group T1 and 155 participants in
group T2 in the first data collection wave)
3. Change in demand occurring in reaction to being made aware of emissions and then being
told that emissions will be offset: 1240 observations (155 participants in group T1 and
155 participants in group T2 in the first data collection wave)
4. Change in demand occurring in reaction to being asked for WTP a repeated time, without
being made aware of emissions: 900 observations (75 participants in Group 4 observed
twice and 75 participants in group 5 in the second data collection wave)
5. Change in demand occurring in reaction to being shown emission labels, without previously
being made aware of emissions: 900 observations (150 participants in group 6 and 75
participants in group 5 in the second data collection wave)
6. Change in demand occurring in reaction to being told that emissions will be offset: 600
observations (150 participants in group 6 in the second data collection wave)
In the main analysis, I am interested in whether these treatment effects significantly differ and in
how this interacts with the greenhouse gas emissions caused by the meal in question. Treatment
effects (3) and (6) can be pooled to provide insights to the structural model for decision making
under reduced environmental concerns, while treatment effects (1),(2),(4) and (5) together form
a two by two design varying across the dimensions of awareness (by guessing emissions) and
information (through emission labels).
As an additional analysis, the effects of providing emission labels on WTP can also be analyzed
using only data collected in the second data collection wave, comparing only treatment effects
(4) and (5) in the list above.

WTP, guess and survey data can be pooled with data from the first data collection wave to add
to the following secondary analyses (already described in the original PAP):
• WTP for carbon labels without and with having experienced carbon labels previously (as
elicited in step 8) suggests the (expected) effect on welfare of being provided with these
labels. The carbon labels tested in the experiment were designed together with Bonn’s
student restaurant and the student restaurant is considering implementing these labels on
a large scale in the future.
• Participant’s guesses for the emissions attributable to meals can be tested for their accuracy.
Camilleri et al. (2019) found that people are insufficiently sensitive to the magnitude
of differences in emissions between food items. I expect consumers to overestimate the
emissions of low- and underestimate the emissions of high-carbon meals.
• The data gathered in step 2 can be used to construct a demand curve for each meal,
allowing to evaluate the effect a carbon tax would have. Thus, one can compare the
effectiveness of carbon labels versus carbon tax as policy instruments.
• In step 10 of the experiment, participants are asked for their approval of (1) the introduction
of carbon labels and (2) the introduction of a carbon tax in the student restaurant. I will
examine whether approval differs between treatment groups. Further, these answers can
be used as a check on the WTP which participants indicate for being shown emissions
information.
• Suggestive within-subject estimates of treatment effects can be constructed by comparing
step 4 and step 5 WTP for a given meal of a given subject with her baseline WTP for
the meal. This allows for some heterogeneity analysis. The effectiveness of the label
might differ depending on (1) subjects’ education, (2) subjects’ income, (3) subjects’
environmental attitude, (4) subjects’ degree of self control in eating. The same factors
might influence subjects’ WTP for being shown the label.
• One might argue that participants shown emissions labels use these labels to infer nutritional
characteristics of the meal. To check whether this is the case, I have participants
guess the calories attributable to meals in step 6 of the first data collection wave/ step 7
of the second data collection wave. One group of participants is not shown emission labels
for this guess, while other participants are shown the emission labels. If it is the case that
participants infer nutritional information from emissions labels, the guesses made by the
two groups should systematically differ.
Similarly, data from the second data collection wave can also be included in the estimation of
the structural model.

The experiment studies incentivized choices between different restaurant meal options. The experimental conditions vary in the information which is provided to participants at the moment of choice.

Experimental sessions are planned for the end of October and beginning of November 2021.
In each session, participants first fill out a survey online (45 minutes duration) and then pick
up their payment in person on the same day and no later than 2 pm. In addition to their
payment, participants receive a meal for immediate consumption when picking up their payment.
Participants can only sign up for one of the sessions.
The timing of the survey and pick-up of payment was chosen so as to mimic the usual process of
choosing a lunch meal as far as possible. Due to COVID-19 regulations, the survey is conducted
online instead of in-person. In some parts of the survey, participants make incentivized guesses:
of the greenhouse gas emissions and calories attributable to each meal and of entirely unrelated
issues (as a time-filling task to proxy for the greenhouse gas emissions guessing task). These
guessing tasks are restricted to 60 seconds so that participants are not able to search for solutions
online.
Participants are randomly sorted into either group 4 (total of 75 participants), group 5 (total
of 75 participants) or group 6 (total of 150 participants). These treatment conditions add to
those observed in the first data collection wave (groups T1, T2 and C). Depending on the
treatment group participants are assigned to, the information conditions under which they make
consumption decisions in the core part of the survey differ.
In the course of the survey, participants’ willingness to pay (WTP) for various restaurant meals
is elicited under different information conditions. In one information condition, participants are
shown the greenhouse gas emissions caused by each meal. These emission values are calculated
using the Eaternity Institute (2020) database. I purchased an Eaternity personal license, and
Eaternity has confirmed that I may use this license to calculate values for the experiment.
The experiment procedure is:
1. Questions on demographic information, allergies, eating preferences, current hunger level.
2. WTP elicitation for meals A,B, C and D.
3. A time filling task in which participants make incentivized guesses on completely unrelated
issues. This task is to proxy for the greenhouse gas emissions guessing task experienced by
groups T1, T2 and C in the first data collection wave, without drawing any attention to
the issue of greenhouse gas emissions. For example, participants are asked to estimate how
much world population increased over the past 20 years, how long a certain running route
in Bonn is and how many yellow cards were shown in the last global football tournaments.
For incentivization and to keep the same protocol as in the emissions guessing task,
additional e0,10 are added to participant’s payment for every guess within 30% of the
true value. Each guess is restricted to 60 seconds.
4. Repeated WTP elicitation for meals A, B, C and D.
• Groups 4 and 5 repeat the previous baseline WTP elicitation.
2
• Group 6 is now shown emission labels for each meal.
5. Repeated WTP elicitation for A, B, C and D.
• Group 4 repeats the previous baseline WTP elicitation.
• Group 5 is now shown emission labels for each meal.
• Group 6 is told that the emissions attributable to the meal chosen will be offset.
6. Group 4 (who has not yet seen any emission labels) guesses the greenhouse gas emissions
caused by eleven different meals, analogous to the emissions guessing task performed by
groups T1, T2 and C in the first data collection wave. However, one additional meal was
added to the guessing task to provide further insights. For incentivization, additional
Euro 0,10 are added to participants’ payment for every guess within 30% of the true value.
Each guess is restricted to 60 seconds.
7. Incentivized guess of the calories attributable to meals A, B, C, D and the cheese sandwich.
Groups 4 and 6 are shown emission labels in this procedure, while group 4 is not.
8. WTP elicitation for receiving emissions information for meals E, F and G.
9. WTP elicitation for meals E, F and G, with information conditions depending on the
previous decision.
10. Participants answer questions on attitudes towards the environment and psychological
traits such as self control in eating. Further, they are asked how much they would support
the introduction of (1) carbon labels or (2) a carbon tax in the student restaurant.
In steps 2, 4, 5 and 9 of the survey, participants make a total of 15 consumption decisions. Each
decision is a choice between receiving a cheese sandwich or a warm meal. This warm meal is a
typical student restaurant meal, and the meals which are handed out to experiment participants
after completing the experiment are in fact prepared by Bonn’s student restaurant. The cheese
sandwich is also prepared by the student restaurant and is a typical cheese sandwich (bread roll,
slices of cheese and some lettuce garnish).
Regardless of the decisions participants make in the survey, they always receive one meal at
pay-out (i.e. cheese sandwich or warm meal). This mimics usual meal choice: the alternative to
not eating a certain meal is not "not eating", but eating something else. The WTP captured
for a certain meal is thus relative to the participants’ WTP for a cheese sandwich, as it is the
participant’s WTP to receive the meal instead of the cheese sandwich. If a participant prefers
the cheese sandwich, this is interpreted as negative relative WTP for receiving the meal. As the
main object of interest in this study is the change in WTP for meals which is induced by the
treatments, it is secondary whether absolute or relative WTP values are captured and analyzed.
In each of the 15 decisions, participants first state whether they prefer receiving the cheese
sandwich or the warm meal at payout, and then state the maximum amount they are willing to
pay to exchange the two options if they are handed their less-preferred option. Participants are
incentivized to respond truthfully, since one of these decisions is in fact implemented. For this
decision, with 50% probability, a participant is handed their preferred option for free. With
50% probability, she is first allocated the less-preferred option, and receives her preferred option only if her WTP lies above a price which is randomly drawn from the interval (0,3), where each
value in 5-cent steps is equally likely. If her WTP lies above the price drawn, the drawn price is
automatically deducted from the participant payment. If her WTP lies below the price drawn,
she receives her less-preferred meal and no amount is deducted.
For each step, the order in which meals are shown to participants is randomized, i.e. there
is randomization across meals A, B, C and D, there is randomization across the incentivized
emission guesses and there is randomization across meals E, F and G. Further, one aspect of
the layout of the design decision - whether the warm meal or the sandwich is shown on the left
or right part of the screen - will differ across experimental sessions to ensure that results are not
driven by this feature.
Which decision is relevant for pay-out is partly pre-determined for logistic reasons, but not
known to the participants. Great care was taken to ensure that participants are not able to guess
which of the decisions is relevant for pay-out. For each participant, there are a total of seven
meals playing a role in her 15 payout decisions. These seven meals differ depending on whether
the participant is vegetarian or not. On each day, the meal which is relevant for payout is the
same across non-vegetarian participants and the same across vegetarian participants. However,
the relevant meal differs across days. It is thus not possible for participants to potentially learn
from experiment participants from previous days which of the meals is relevant. Further, all
meals asked for in the experiment are typical student restaurant meals and are regularly offered
by the student restaurant in Bonn, so that participants should not be inferring that one of the
meals is unlikely to be relevant.
The meals for which WTP is elicited in the second data collection wave are identical to those of
the first data collection wave. Correspondingly, the same meals will be relevant for pay-out as
in the first data collection wave, of course again split across days in such a way that it is not
possible to predict which meal will be relevant for payout on a given day.
For the WTP elicitation for meals E, F and G in step 8 of the survey, participants have the
opportunity to purchase emissions information for these meals. Participants decide whether
they prefer the information to be shown, and indicate a WTP for their preferred display option.
With 50% probability, a participant’s preferred display option is implemented for free. With
50% probability, she is first allocated the less-preferred display option, and receives her preferred
option only if her WTP lies above a price which is randomly drawn. The price drawn for this
information is only deducted from participants’ payment if one of the final three decisions is the
decision relevant for payout. Under these information conditions, the WTP for meals E, F and
G is elicited.

Participants are randomly sorted into treatment groups.

Randomization at the level of the individual

No

300 individuals

300 individuals

75 participants in group 4,
75 participants in group 5,
150 participants in group 6

Details in attached PDF. The standard error of the effect size is assumed to be 0.0217. With 300 participants in the experiment, MDE is 0.0583 (9.5%) With 200 participants in the experiment, MDE is 0.0714 (11.6%)