Food Policy 66 (2017) 1224

Contents lists available at ScienceDirect

Food Policy
journal homepage: www.elsevier.com/locate/foodpol

Two experiments in one: How accounting for context matters

for welfare estimates
Stphan Marette a,, Christophe Martin b, Fabienne Bouillot b
a
conomie Publique, INRA, Universit Paris-Saclay, F-78850 Grignon, France
b
INRA, UMR1324 Centre des Sciences du Got et de lAlimentation, Dijon, France

a r t i c l e i n f o a b s t r a c t

Article history: By combining two different types of experiments in one experimental session, this paper aims at under-
Received 20 February 2015 standing how different contexts may influence participants choices. This paper focuses on one hybrid
Received in revised form 7 October 2016 experimental session that mixed one voluntary contributions mechanism (VCM), influencing the indem-
Accepted 17 November 2016
nity received by participants, and one mechanism eliciting willingness to pay (WTP) for milk bottles with
public and private attributes. The VCM shows relatively high levels of contributions that are mainly influ-
enced by the positive expectations of participants about the average group contribution, rather than by
JEL classification:
the variations in the design of this mechanism and the period of experiments. The WTP for milk bottles
C92
H41
are particularly sensitive to the order of mechanisms and to the period of experiments. Conversely, the
H42 WTP differences between milk bottles for a given round of information are invariant across the order
of mechanisms and the period of experiments. For each bottle, the variations of WTP coming from the
Keywords: messages about private and public attributes are also stable over the order of mechanisms and the period
Experiments of experiments. This confers validity to experiments for measuring WTP for public and private attributes
Public goods related to food. In other words, these variations of WTP contribute to welfare estimates and are useful to
Private goods
evaluate market regulations focusing on public and/or private attributes.
Validity
2016 Elsevier Ltd. All rights reserved.
Welfare
Food

1. Introduction economists, many factors may sway the determination of prefer-

Willingness to pay (WTP) for food are intensively used for esti-
mating the impact of new product adoption or quality improve-
ment, as well as for cost-benefit analyses. In particular, lab
experiments elicit either WTP for private goods or contributions
to a game with a public good. The lab context is useful for eliciting
well-informed, thoughtful preferences. However, this advantage is
hobbled by limitations stemming from the artificial environment
and the limited number of products (or payoffs) at stake, while
real-life choices are multi-tasks and imply quick decision under
uncertainty, imperfect information, tasks overload and/or imper-
fect recall (see Levitt and List, 2007).1 As underlined by behavioral
Corresponding author at: UMR Economie Publique, INRA, 78850 Grignon,
France.
E-mail address: marette@agroparistech.fr (S. Marette).
1
These criticisms explain the economists efforts to turn to field experiments for
understanding social choices under real-life contexts. However, Falk and Heckman
(2009) mention that testing some complex theories or algorithms with field
experiments is notoriously impossible, because people are influenced by too many
incentives and parameters that renders the econometric estimations of field
experiments hard to decipher.
ences, which potentially weakens the credibility of survey/lab
results (see Ariely et al., 2003).
Despite these flaws, the lab allows a tight control of both envi-
ronments and participants actions (see Falk and Heckman, 2009;
Kagel and Roth, 2000). This tight control is particularly precious
for studying the stability of preferences. Extrapolations of experi-
mental results for general explanations regarding behaviors are
only possible, if elicited preferences are relatively stable, both
across different contexts and over a period of time.2
When considering choices coming from the lab for a welfare
analysis related to WTP, this is particularly important to under-
stand what choices are invariant with respect to the design or
the period of experiments. In particular, if choices are relatively
similar with different samples of participants who represent the
same population, extrapolations of experimental results to the
whole population make sense for a consistent welfare analysis
(Roosen and Marette, 2011). This acute question particularly
2
Regarding general foundations in economics, Becker (1976) mentioned that
preferences are assumed not to change substantially over time.

http://dx.doi.org/10.1016/j.foodpol.2016.11.004
0306-9192/ 2016 Elsevier Ltd. All rights reserved.
 S. Marette et al. / Food Policy 66 (2017) 1224
matters with food, for which many experiments/surveys tackled
controversial topics like Genetically Modified Organisms, pesti-
cides use, animal cloning, meat substitutes, animal welfare, food
safety, food security, food wasting, protection of resources or sus-
tainability of crops. . .
This paper aims at precisely understanding how different
designs and/or periods of time may influence or not participants
choices in the lab. This paper focuses on two successive experi-
ments (or mechanisms) run in one session. Each hybrid experi-
mental session combined one VCM influencing the indemnity
received by participants, and one elicitation mechanism measuring
WTP for private goods offered to participants. The VCM was a sim-
plified game of voluntary provision to a threshold public good,
leading to the possibility to increase the indemnity that partici-
pants could receive, if the percentage of contributors choosing to
invest was higher than a contribution threshold. For eliciting
WTP, we chose milk bottles from cows fed a diet without linseed
and alternative milk bottles from cows fed a diet with linseed. Pur-
chasing milk from cows fed a diet with linseed contributes to
reduce methane emissions from cows (namely, a public attribute)
and increase content in omega-3 polyunsaturated fatty acids in
milk.3 Successive WTP for both products were elicited after succes-
sive messages on the impact of linseed.
We conducted the experiments in the same tasting room in
Dijon, France, in April and November 2013. Different sessions
allowed minor variations regarding some parameters of the VCM
(threshold levels and/or money back conditions) and the order of
the two different mechanisms, namely the VCM preceding the
WTP elicitation and vice versa. With our between-subject design,
the participants attending the April sessions were different from
participants attending the November sessions, even if the ex ante
criteria of recruitments for both periods were strictly similar.
New participants in November allow us to guarantee the novelty
of VCM and the novelty of messages on linseed preceding succes-
sive WTP elicitation. A priori, with samples of participants who rep-
resent the same population of a city like Dijon, the order of two
mechanisms and/or the periods of sessions conducted a few
months apart should not influence the choices, if preferences are
stable.
Results are the following. The VCM shows a relatively high level
of contributions that is mainly influenced by the positive expecta-
tions of participants about the average group contribution, rather
than by variations in the design of this mechanism. The contribu-
tions to the VCM are invariant with the order of mechanisms and
the period. The WTP for milk bottles are particularly sensitive to
the orders of mechanisms and the periods of experiments. Con-
versely, the WTP differences between milk bottles for a given
round of information are invariant across the order of mechanisms
and the period of experiments. For each bottle, the variations of
WTP coming from the messages about private and public attributes
are also stable over the order of mechanisms and the period of
experiments. When compared to regulatory costs, these variations
of WTP can credibly contribute to evaluate market regulations
focusing on public and/or private attributes. Eventually, economet-
ric estimations show some positive links between the participants
contribution to the VCM and the WTP variations, coming from
information about the public attribute related to greenhouse gases
and linseed in milk, even this link is weak and not systematic.
By combining two different mechanisms in one unique and
hybrid session, this paper differs from previous papers, only using
3
According the Kotchens definition, milk can be seen as an impure public good
generating numerous private attributes like the taste, the nutrition, the long-term
health impact (. . .), and a few environmental public attributes related to greenhouse
gas emissions and other environmental pollution coming from cows (see Kotchen,
2005).
13
VCM for public goods, or only focusing on WTP for private goods.
The possible influences between the two types of different mecha-
nisms were overlooked by previous papers. To the best of our
knowledge, this paper contains the first attempt to combine one
VCM and one WTP elicitation mechanism for a private good with
public and private attributes.
Our paper is mainly related to papers studying consistency and
stability of preferences. Previous studies suggest opposite results
when stability of preferences is studied. Regarding public goods
experiments, Volk et al. (2012) find that cooperation preferences
are very stable at the aggregate level, and, to a smaller degree, at
the individual level.4 However, Brosig et al. (2007) found the oppo-
site result, since they observe a decrease of cooperative behavior
over time. Regarding valuation of environmental attributes with
methods based on hypothetical choices, Loomis (1989), Bliem et al.
(2012) and Brouwer (2012) tend to show stability of preferences,
while Liebe et al. (2012) and Schaafsma et al. (2014) contest this sta-
bility of preferences. Regarding the WTP for private goods, Ariely
et al. (2003) show that WTP, rather than being stable, are highly sen-
sitive to irrelevant influences and surprising anchoring..Ariely et al.
(2003) also underscore that WTP differences between two different
bottles of wine vary a lot across subgroups of participants. Alfnes
et al. (2011) also reject preference stability of WTP for private goods
at the individual level, but not at the aggregate market level.
Our paper differs from these previous contributions by gather-
ing all the previous dimensions in one experiment, namely the
public good aspect with the VCM, the environmental/public attri-
bute with the reduction of methane emissions, and the WTP for
private goods with the bottles of milk.5 We also pay attention to
an overlooked issue, namely the questions of the stability of WTP
differences between products for a given round, and the variations
of WTP for each product coming from information about public or
private attributes.6 In our paper, these variations of WTP are invari-
ant across the design and the period of experiments, which differs
from Ariely et al. (2003) underlining large WTP differences between
products, only for a single round of WTP determination. These ques-
tions were not studied by Alfnes et al. (2011).7
Our paper is also related to previous experiments studying pub-
lic goods, including the threshold public goods (see for instance
Cadsby and Maynes, 1999; Croson and Marks, 2000). Coats et al.
(2009) show the positive impact of the refund of the investment
if the threshold is not reached, a result that does not hold in our
VCM in November. In our paper, the novelty also comes from the
fact that the order of mechanisms and the period of the experiment
do not influence the participants contribution to the VCM.
This paper is organized as following. Both experiments are
detailed in the next section. The third and fourth sections
4
Andersen et al. (2008) find stability in risk attitudes over a 17-month span.
5
Our paper also differs from previous studies on multiple public goods analyzed
with several VCM. In particular, Moir (2006) and Bernasconi et al. (2009) show that
contributions increase when public goods are split into two identical public goods
(see also Falk et al., 2013; McCarter et al., 2013; Corazzini et al., 2013) Our approach
differs by showing a weak link between the participants contributions to the VCM
and the variation of WTP for the public attribute related to greenhouse gases and
linseed in milk. As these two mechanisms are not similar, participants disconnect
their contribution to each mechanism.
6
Our study also contributes to the literature on global warming and climate
change, by focusing on consumers choices between two foods related to climate
mitigation, via a reduction in methane emissions. This differs from a recent public-
good experiment made by Hasson et al. (2010), in which mitigation of greenhouse
gases is viewed as a public good and adaptation to climate change is viewed as a
private good, but without any reference to existing products sold in supermarkets.
Conversely, we focus on real bottles of milk.
7
This is also related to the Lusks and Schroeders (2004) result, showing that
marginal WTP for a change in quality/characteristic with food is not statistically
different across hypothetical and real payment settings (see also Taylor et al., 2010).
14 S. Marette et al. / Food Policy 66 (2017) 1224

respectively focus on the VCM and the WTP elicitation mechanism. We now detail the April experiment. Then, we will briefly pre-
The fifth section concludes. sent the November experiment by insisting on the differences with
the April one.

2. The design of experiments 2.2. The April experiment

Each session combined one VCM and one WTP elicitation mech- The sample of participants consisted of 134 people aged
anism. Because of constraints regarding the time and the articula- between 18 and 69 years.
tion between both mechanisms, some simplifying assumptions
were necessary for avoiding long explanations related to complex 2.2.1. The VCM
mechanisms. Each mechanism was conducted separately, and the The VCM consisted in three simplified rounds, with the possibil-
VCM/WTP results determining the indemnity and the sold product ity to increase the indemnity that participants could receive for
were only revealed to participants at the end of each session. purchasing the milk, namely a private good with the public attri-
We conducted the experiments in the tasting room of the Che- bute. Because of time constraints related to the part focusing on
mosens platform (INRA Dijon, France), in multiple one-hour ses- WTP elicitation, we decided to use a discrete choice context (yes
sions in April and November 2013. Many aspects of April and or no), where participants were faced with a choice to invest or
November experiments were strictly similar for allowing compar- not 3 taken from the initial indemnity of 15. A bonus would
ison. This is particularly true for the recruitment of participants. be added to the initial indemnity, only if number of per-session
participants investing the 3 was greater or equal to a contribution
threshold. The payoff structure reflects a prisoner dilemma trade-
2.1. Recruitment of participants
off coming from game theory, such that the dominant strategy is
to invest nothing in the public good if they expect enough other
The sample of participants was randomly selected based on the
participants decided to invest. While this all-or-nothing frame-
quota method and was representative for age groups and socio-
work deviates from the continuous framework of VCMs, this sim-
economic status for the population of Dijon. These ex ante criteria
ple rule is easy to explain in a very limited period of time. The
were rigorously the same for the participants selection in April and
contribution threshold and the indemnity bonus vary across the
November 2013. Table 1 shows relatively minor differences
three rounds. Between rounds of the VCM, we did not reveal any
regarding the socio-demographic characteristics of participants
previous decisions allowing participants to observe and learn,
recruited in April and November 2013. These relatively similar
because such a revelation could add behavioral dynamics by par-
samples of participants allow thorough comparisons between
ticipants. We explained that one round of choices would be ran-
experiments conducted at both periods.
domly selected at the end of the experiment.
For the recruitment, participants were contacted by phone.
Participants were divided in two subgroups G1 or G2 that dif-
They were informed that the experiment would focus on milk con-
fered regarding the contribution threshold, defined by the mini-
sumption and it would last about one hour with a 10 participa-
mum proportion of per-session participants investing 3 for
tion fee. After entering the experimental room, all participants
giving a bonus to all participants. With the subgroup G1, the first
were seated in individual booths in silence during all the session.
round of choices requires a one third contribution threshold, con-
We informed them that we added 5 to the initial compensa-
sisting in reaching 1/3 of per-session participants who agreed to
tion/indemnity for allowing some possible purchases of milk bot-
invest 3 for giving the additional 5 bonus to all participants.
tles. We briefly informed them that a simple game would be also
If at least 1/3 of participants invested 3, the payoff for partici-
held for giving them the opportunity to increase their indemnity.
pants who invested 3 was equal to 15 3 + 5 = 17 and
After these few general explanations, they signed a consent form,
the payoff for participants who did not invest 3 was equal to
and the VCM or WTP elicitation started. Between 6 and 14 partic-
15 + 5 = 20. If less than 1/3 of participants invested 3, the pay-
ipants attended different sessions.
off for all participants was 15. The second round linked to G1 was
similar to the previous one, except that the proportion of partici-
Table 1 pants investing the 3 should be greater or equal to the threshold
Socio-economic characteristics of participants. of 1/2 in order to give the extra bonus. The third round of G1 was
April November the similar to the previous one with the proportion of participants
investing equal to 1/2, except that the bonus given to all partici-
Number of participants 134 113
pants was equal to 4 and 1 was given to a local association in
Participants gender (%)
charge of the environment protection.
Male 47.8 45.1
Female 52.2 54.9 The subgroup G1 was characterized by contribution thresholds
equal to (1/3, 1/2, 1/2) for the three successive rounds. The sub-
Participants age (%)
<30 27.6 31.1 group G2 was similar to the game G1 except for the contribution
[3039] 17.9 19.4 threshold. With G2, the contribution thresholds were equal to
[4049] 19.4 14.2 (1/2, 2/3, 2/3) for the three successive rounds.
[5059] 21.6 21.2
The responses of the subjects were recorded thanks to FIZZ
P60 13.4 14.2
software. At the beginning of the VCM, rules were successively pre-
Participant education (%)
sented by the speaker and recalled on the computer screen. Expla-
No baccalaureate (BAC)a 20.9 17.7
BAC or 2 years after BAC 44.8 41.6
nations were posted for at least 1 min before participants could
More than 2 years after BAC 34.3 40.7 proceed to their choice on the following page. Each time, the
Monthly net income of the household (%)
speaker invited the participants to carefully read the explanations
< 1000 7.5 11.5 before making a choice regarding the 3 investment. This choice
[ 1000 3000[ 56.7 51.3 was followed by a question about the expectation regarding the
[ 3000 5000[ 35.1 32.7 investment choices by other participants. This question asked the
P 5000 0.7 4.4
participants if they thought that the necessary proportion of par-
a
Baccalaureate (BAC) is the French high school diploma. ticipants investing 3 for having the bonus would be reached.
 S. Marette et al. / Food Policy 66 (2017) 1224 15

No result regarding the contributions was revealed after the will have to buy milk (i.e., performance-based financial incentives).
related choices, for avoiding the influence of results on the follow- No reference prices for milk were indicated to participants before
ing rounds provision of public goods or on WTP elicitation, if the the WTP elicitation. It should also be noted that food and milk
WTP was coming after the 3 rounds of provision to public goods. prices observed in stores were quasi-stable in 2013, since the aver-
By randomly selecting one of the three rounds, the result of one age inflation of France in 2013 was equal 0.86%.9 In such a context
VCM determining the indemnity was only revealed at the end of of very low inflation, the difference of WTP for a given product over a
the experiment. period cannot be explained by a change in perception coming from
changes in observed prices between April and November.
Based on different types of information revealed to participants,
2.2.2. WTP elicitation for the public attribute linked to the private good
four rounds of WTP elicitation with the BDM procedure were suc-
The other part of our experiment focused on bottles of milk. We
cessively determined. The two types of milk (with or without lin-
offered 3 L sold by the brand Candia, and called GrandLait, and 3 L
seed for feeding cows) were introduced in the round A with two
sold by the supermarket brand Carrefour since these supermarkets
minutes for observing the bottles. The first round A elicits partici-
are located in Dijon. The choice of these two products reproduced
pants WTP denoted WTPAConv ;i for 3 L of the so-called conventional
conditions of sales very close to real supermarket shelves, where
milk (Carrefour) and WTPALin;i for 3 L of Grand Lait (Candia) for a par-
two national brands usually compete with the supermarket brand.
ticipant i (with i = 1, . . ., N = 134).
The choice of these two products was also justified by differ-
Regarding the following rounds j e {B, C, D}, successive types of
ences regarding greenhouse gas emissions that are a public bad.
precise information were then communicated. Messages were
Because of enteric fermentation, ruminants are emitting a lot of
made from a review of agronomic and nutritional literatures and
methane that is a greenhouse gas. Adding oilseed lipid supple-
they only revealed benefits coming from linseed. Messages were
ments from linseed to the diet of dairy cows significantly reduces
relatively short, because some previous works underline that a
the production of methane (Martin et al., 2008). Linseed appears
short message is more efficient compared to long messages with
as the most efficient supplement leading to configurations with a
complex information (Wansink et al., 2004). Details about the
20% or 30% reduction in methane emissions (Nguyen et al.,
complete messages and the order of information are detailed in
2012). Purchasing milk from cows fed a diet with linseed con-
Appendix A.
tributes to reduce a public bad related to greenhouse gas. Beyond
the reduction of greenhouse gas emissions, linseed in dairy cows
(1) In round B, a first type of message revealed environmental
diet also leads to milk with a higher content in omega-3 polyunsat-
information about the impact of linseed on the reduction
urated fatty acids protecting the consumers cardiovascular sys-
of methane emissions.
tem.8 Successive messages on linseed and their impacts on the
(2) In round C, a second message explains the public dimension
environment were delivered to participants, and WTP for both prod-
of the methane emissions and the negative (respectively
ucts were elicited after each messages.
positive) externality linked to the purchase of the Carrefour
The milk with the Carrefour brand does not use linseed for feed-
(respectively Grandlait) milk.
ing cows. The choice of the Candia brand is motivated by its use of
(3) In round D, a last message revealed health information about
linseed for feeding cows leading to a label Filire Nutrition, Bleu-
benefits linked to omega-3, by underlining their positive
Blanc-Coeur posted on the bottle and indicating a high-content in
impact for reducing cardio-vascular diseases.
omega 3. The size of this label Filire Nutrition, Bleu-Blanc-Coeur
posted on the bottle is relatively small (length: 2 cm, width:
1.5 cm). A few sentences posted on the bottle and linked to this
2.2.3. The experimental design
label mention an improved diet for cows leading to more omega-
At the beginning of the experiment, participants received gen-
3, but the use of linseed is not explicitly indicated. The environ-
eral instructions and signed a consent form. In the initial explana-
mental impact was not indicated on this bottle. Three other labels
tions of both experiments, we insisted on the anonymity of replies,
of a similar size are posted on this bottle (see note 2). In 2011, the
since participants were only identified with a number during the
market share in volume of milk with this label was 5% (see Blanc-
experiment.
Bleu-Coeur Association, 2012).
We conducted the experiment in four treatments, varying the
To elicit participants WTP, our experiment used the BDM pro-
VCMs (G1 or G2) and changing the order between the VCM and
cedure (Becker et al., 1964). Successive items of information were
the WTP elicitations. The left part of Fig. 1 shows the timelines of
revealed to participants, and a WTP was elicited after each message
both subgroups with different orders regarding mechanisms.
for both types of milk. The exact question was as follows: What is
At the end of the experiment, after the round D, the indemnity
the maximum price you are willing to pay for the following bottle
was determined by the random selection of one of the three VCM
of milk? The question was repeated for each product. We insisted
choices and the revelation of the contribution to VCM for knowing
on the freedom to report any WTP, including zero, if they did not
whether or not the threshold was reached. Then a last round E was
want to purchase milk.
organized for knowing WTP after the revelation of the indemnity
The part with the private good started with a trial round to
(see the bottom of Fig. 1).
explain the BDM mechanism. Simulations with a small carrot
Participants completed an exit questionnaire on related issues
helped participants understand the mechanism. The possibility of
and socio-demographic characteristics. The experiment concluded
zero bids in the BDM procedure was carefully explained, as well
by randomly selecting one type of milk (with or without linseed for
as the possible use of the 15 or more (depending on the bonus
feeding cows) and one of the fifth elicited WTP, which were used to
related to the VCM) for making the real purchase. We also made
determine whether participants take the bottles away with them.
clear that one of the elicited WTP would be randomly selected at
A price between 0.1 and 5 was also drawn at random, and pur-
the end of this experiment for determining whether participants
chasing choices were enforced. If the selected WTP was smaller
than the randomly drawn price, the participant received the 15
8
Despite these advantages coming from linseed, 95% of milk bottles sold in France indemnity or more (depending on the bonus related to the VCM).
in 2011 offer milk from cows fed a diet without linseed, namely with corn or other
fodders. The absence of linseed for 95% of bottles comes from (1) the relatively high
9
price of linseed compared to other types of fodder for cows and (2) the non- See http://www.inflation.eu/inflation-rates/france/historic-inflation/cpi-infla-
internalization of different advantages of linseed by consumers. tion-france-2013.aspx (accessed 01.06.2016).
16 S. Marette et al. / Food Policy 66 (2017) 1224

April November

Initial instructions Initial instructions

Explanation BDM Explanation VCM Explanation VCM

WTP elicitations VCM VCM

4 rounds: A, B, C, D 3 rounds of choices 3 rounds of choices
WTPCONV & WTPLIN

Explanation BDM Explanation BDM

Explanation VCM

WTP elicitations WTP elicitations

VCM 4 rounds: A, B, C, D 4 rounds: A, B, C, D
3 rounds of choices WTPCONV & WTPLIN WTPCONV & WTPLIN

Random selection of VMC choice Random selection of VMC choice

for the indemnity for the indemnity

WTP elicitations, Round E WTP elicitations, Round E

WTPCONV & WTPLIN WTPCONV & WTPLIN

Exit questionnaire Exit questionnaire

Random selection of purchase price Random selection of purchase price
Product purchase Product purchase

Fig. 1. Timeline.

If the WTP was equal to or higher than the price, the compensation
was equal to 15 or more (depending on the bonus related to the
selected VCM before round E) less the price randomly drawn, and
the participant received the milk bottles.
The top of Table 2 describes the number N of attendants in each
subgroup of April.
2.3. The November experiment
The sample consisted of a new group of 113 people aged
between 18 and 69 years recruited for multiple one-hour sessions
by following the same recruitment criteria used in April 2013.
Experimental conditions and mechanisms described above
were the same. Compared to April, the four changes mainly con-
cern some points of the VCM in order to study robustness of results
under various designs. The four changes are the following.
(1) The order between mechanisms was the same throughout
the November experiment with the VCM always preceding
the WTP elicitation mechanism as shown on the right of
Fig. 1.
(2) The contribution thresholds were equal to (1/5, 4/5, 4/5) for
the three successive rounds of choices and for all the
participants.
(3) In one subgroup, G4, participants were losing their invest-
ment of 3 (absence of money back or refund), if they
decided to contribute and if the contribution threshold
linked to the round was not reached. In other words, if a pro-
portion of participants invested 3 was lower than the con-
tribution threshold, the payoff for all participants was
15 3 = 12. For the subgroup G3, the mechanism was
similar to the April experiment with no loss if the threshold
was not reached.
(4) The round #3 changed compared to the April experiment.
The gift given to an environmental association of 1 at
round #3 if the threshold was reached was withdrawn,
and the bonus given to all participants was equal to 5
(as in previous rounds). For this third choice of the VCM,
information was introduced about the previous choice
and the proportion of contributors. The message was the
following: At the previous round, the proportion of partic-
ipants who contribute was higher (respectively lower) than
the threshold equal to 80%. We alternate the positive and
negative information between successive sessions what-
ever the results in round #2 of the session, for precisely
controlling both positive and negative information across
groups. Table 2 describes the number of attendants in each
subgroup.
 S. Marette et al. / Food Policy 66 (2017) 1224 17

Table 2
Experimental design.

April G1 1/3, 1/2, G2 1/2, 2/3, 2/3

Different thresholds
VCM WTP N = 41 N = 25
WTP VCM N = 30 N = 38
November G3 Without loss G4 With loss
Possible loss if the threshold is not satisfied with the VCM
VCM, Round 3: Negative info N = 26 N = 29
VCM, Round 3: Positive info N = 29 N = 29

Note: N is the number of participants per subgroups.

tial willingness to cooperate outweighing the incentive to deviate

3. Results related to the VCM
Regarding the VCM, the average contributions are summarized
by subgroups in Fig. 2, in which the contributions thresholds speci-
fic to the 3 successive rounds are represented on the X-axis and the
percentages of contributors investing 3 are represented on the Y-
axis. Different subgroups are represented on the fourth charts for
illustrating the diversity of mechanisms. Fig. 2 shows how the
average contribution changes with different contribution
thresholds.
For the first contribution of participants, the average contribu-
tion, represented on the left of each chart, is higher than the corre-
sponding threshold levels indicated on the X-axis. It means that, in
average, the indemnity increased when the first round was
selected before round E. The four charts of Fig. 2 underscore an ini-
for having a higher payment.
For understanding the impact of thresholds on contributions,
we particularly test the significance of differences of contributions
between rounds, by using the Wilcoxon test for paired samples,
and with D indicating the significant differences at the 1%, 5%,
and 10% levels. The shifts of average contributions between the
first round with a given threshold and the second round with a
higher threshold in the four charts of Fig. 2 inform us about the
impact of the variation in the contribution thresholds within each
subgroup. Except for the subgroup G3, the changes in the contribu-
tions coming from different thresholds are not statistically signifi-
cant at the 5% level (this view will be mitigated in Table 4).
The variations between the second round and the third round
inform us about some variations in the design of the VCM, since
the contribution thresholds are the same. The charts at the top of

Fig. 2. Percentage of contributors in the VCM (average by subgroups). Note: N participants are taken into account. D denotes significant difference at 10%, D denotes
significant difference at 5% and D denotes significant difference at 1% as tested by the Wilcoxon test for comparing paired sample choices regarding contributions.
18 S. Marette et al. / Food Policy 66 (2017) 1224

Table 3 shows the significant, but complex effects coming from the addi-
Across sample comparison (Mann-Whitney-U test) for the contribution to VCM from tional information given before the third choice.
different subgroups.
Table 3 confirms the weak impact of threshold differences by
April Impact of the Impact of the VCM/WTP focusing on the across sample comparisons between subgroups
contribution threshold order with Mann-Whitney-U test. In particular, differences between sub-
G1 (N = 71) versus G2 VCM first (N = 66)
(N = 63) versus WTP first
groups of the April experiment are not statistically significant at
(N = 68) 10%. The top and left part of Table 3 focusing on the comparison
between groups G1 (N = 71) versus G2 (N = 63) with different
z-value p-value z-value p-value
thresholds confirm the absence of significant impact of different
Choice 1 1.029 0.303 0.741 0.458
thresholds for the April experiment. Moreover, Table 3 shows that
Choice 2 0.549 0.582 1.629 0.103
Choice 3 0.289 0.772 0.066 0.947 the order between the two mechanisms (VCM and WTP) does not
significantly impact the decision of positively contributing. Even-
November Impact of the possible Impact of information
loss in 3d round VCM tually, the presence/absence of a loss linked to the investment
G3 (N = 55) versus G4 Neg. info (N = 55) when the threshold is not reached does not significantly impact
(N = 58) versus Pos. info (N = 58) the decision of positively contributing.
z-value p-value z-value p-value Econometric estimations of contribution presented in Table 4
Choice 1 1.232 0.218
confirm the previous results. Table 4 underscores the estimation
Choice 2 1.425 0.154 for April in the first column, the estimations for November in the
Choice 3 0.686 0.492 1.440 0.150 second and third columns, and the estimation for the combined
Note: N is the number of participants in subgroups. Across sample comparison by
(stacked) data from April and November in the last column. Given
Mann-Whitney-U test. H0: frequencies for a group = frequencies for the other group. that each participant i made three binary choices (Yes or No), there
denotes rejection of H0 at 10% significance level. denotes rejection of H0 at 5% should be some correlation across data points relating to contribu-
significance level. denotes rejection of H0 at 1% significance level. tion, which leads us to use the random effects Logit estimator. We
tested for the influence of many variables on contribution, and we
re-run regressions without the variables characterized by a coeffi-
Fig. 2 shows the ambiguous impact coming from the gift to the cient that was statistically non-significant. The only exception for
environmental association for the third choice, with one significant which we reported the non-significant coefficients in the model
effect in G2, and non-significant effects in G1. The bottom of Fig. 2 (1) are the ones directly related to the design of VCM. For

Table 4
Decision to contribute to the public good and invest 3 with the VCM.

Dependent variable Decision to contribute at rounds 1, 2 and 3

Specification Logit random effects
April November Combineda
Model (1) (1) (2) (1)
Expectation regarding the contribution by others, dummy = 1 if contribution by others > threshold 1.945 3.811 3.838 2.768
(0.392) (0.583) (0.578) (0.332)
Contribution threshold (in percentage) 0.001 0.038 0.038 0.026
(0.013) (0.009) (0.009) (0.006)
Order of mechanisms (April, Combined), dummy = 1 if VCM before WTP for milk 0.603 0.874
(0.444) (0.495)
Gift to an environmental association (April, Comb.), dummy = 1 if round #3 in April 0.695 0.603
(0.339) (0.328)
Loss if overall contribution is lower than the threshold (Novem., Combin.), dummy = 1 if loss 0.160 0.232
(0.529) (0.534)
Information about previous contribution and revealed before choice #3 (Novem. Combin.), 0.847 1.001 0.779
dummy = 1 if choice #3 in November (0.572) (0.464) (0.409)
Negative information  information before choice #3 (November only), dummy = 1 if negative information 0.270
(0.714)
Period of the experiment (Combined only), dummy = 1 if November 0.696
(0.544)

WTPBConv,iWTPAConv,i, premium coming from the message on methane for convent. Milk 2.549 1.050
(0.952) (0.597)
WTPAConv,i, WTP for the conventional milk at round A 0.588 0.519
(0.258) (0.228)
WTPCLin,i WTPBLin,i premium coming from the message on the externality of linseed 4.007 3.878
(2.175) (2.146)
Education (from 1 for low-level to 5 for top level)b 0.462
(0.164)
Ageb 0.071 0.070 0.041
(0.020) (0.019) (0.013)
Constant 7.311 7.374 5.617
(1.285) (1.261) (0.917)
Stand. Devi. 2.755 2.844 2.827 2.957
(0.480) (0.581) (0.573) (0.373)
Observations N = 402 N = 339 N = 339 N = 741
Log likelihood 211.851 156.700 156.814 377.696

Note: : significant at 1%; : significant at 5%; : significant at 10%. Standard errors in parentheses.
a
Data from April and November are stacked.
b
Question from the exit questionnaire.
 S. Marette et al. / Food Policy 66 (2017) 1224
November only, the model (2) gave alternative regressions by
withdrawing these variables with non-significant coefficients.
Table 4 confirms the absence of significant influence of vari-
ables related to the VCM design, more particularly the order
between the two mechanisms (VCM and WTP) in April and the
presence/absence of a loss linked to the investment when the
threshold is not reached in November. In the last column with
the combined data, these two variables are either not significant
for the loss, or only weakly significant at 10% for the order of mech-
anisms. The effect of the contribution threshold is only significant
in November with a coefficient equal to 0.038, which may be
explained by the higher difference between thresholds in Novem-
ber (4/51/5) compared to the differences in April (1/21/3 or 2/3
1/2). In other words, the average contribution significantly
increases, only if the shift in the contribution threshold is high
enough, as it is the case in the November experiment. This effect
is also significant in the last column with the combined data, con-
firming the importance of large threshold variation of November.
Table 4 shows the significant impact related to the gift to the
environmental association in the April sessions at the round #3.
In November, the revealed information about the average contribu-
tion in the previous round has a negative impact on contributions
and is only significant in the model (2) with a coefficient equal to
1.001. The positive or negative message regarding the previous
contribution did not matter as underlined in the model (1) of
November.
In the first line of Table 4, the expectation about the proportion
of others participants likely to invest and complying with the con-
tribution threshold significantly explains part of the own partici-
pants contribution to the public good. Recall that this question
was asked after each round of choice made by each participant.
The more participants expect the others to give, the more partici-
pants are likely to give. For each experiment and for the combined
data of the last column, the estimated coefficients are positive and
significant at the 1% level, showing a significant influence of expec-
tations regarding the choices realized by other participants. The
VCM underlines the importance of conditional cooperators
whose contributions to the public good are positively correlated
Fig. 3. Average WTP for 3 L of milk (). Note: N participants are taken into account. D denotes significant difference at 10%, D denotes significant difference at 5% and D
denotes significant difference at 1% as tested by the Wilcoxon test for comparing paired sample choices WTPA and WTPB, WTPB and WTPC, WTPC and WTPD for the different
stages of information revelation.
19
with their expectations about the average group contribution.
The education level negatively influences the contribution in April
and the age positively influences this contribution in November.
Table 4 exhibits links between the VCM and WTP elicitations
related to the milk for both experiments realized in April and
November. The April experiment shows a significant and positive
influence of the WTP for the conventional milk at round A on the
contribution to the VCM with a coefficient equal to 0.597 (other
rounds or WTP of milk from cows fed a diet with linseed have a sig-
nificant impact but leads to less precise estimations). No significant
effect of WTP for a product on the contribution was found in
November.
Table 4 also shows a statistical link between the contribution to
the VCM and the variation of WTP coming from information about
the public good, even if the attributes differ. The variation of WTP,
WTPBConv ;i WTPAConv ;i , coming from the message on methane for
the conventional milk sways the contribution in the April experi-
ment and with the combined data (in the last column). For contri-
butions in November, the variation of WTP, WTPCLin,i WTPBLin,i,
coming from the message on the externality for the milk with lin-
seed sways the November VCM. The variation of WTP for the
omega 3 content in milk that is a private attribute has no signifi-
cant impact on the contribution to the VCM.
Table 4 shows that there is a significant link between the con-
tribution to the VCM and the variation of WTP coming from infor-
mation about public good, even if the attributes differ, which
suggests a weak influence across the periods. The premium,
WTPBConv ;i WTPAConv ;i , coming from the message on methane for
the conventional milk sways the contribution to the VCM in April,
while the premium, WTPCLin,i WTPBLin,i, coming from the message
on the externality for the milk with linseed sways the contribution
to the VCM in November.
Eventually, in the last column of Table 4, data from April and
November were combined, and we added a dummy variable
related to the Period of the experiment and equal to one for obser-
vations of November. This variable Period of the experiment is not
statistically significant, underlining the stability of contributions
over both periods. For concluding, this section has shown that
20 S. Marette et al. / Food Policy 66 (2017) 1224
Table 5
Across Sample Comparison (Mann-Whitney-U test) for WTP linked to different subgroups in the April experiment.
Impact of the order of mechanism
WTP VCM, April, N = 68 versus VCM WTP, N = 66, April
Round A z-value p-value
WTPAConv,i 2.712 0.007
WTPALin,i 1.939 0.052
Note: N is the number of participants in subgroups. Across sample comparison by Mann-Whitney-U test. H0: frequencies for a group = frequencies for the other group.
denotes rejection of H0 at 10% significance level. denotes rejection of H0 at 5% significance level. denotes rejection of H0 at 1% significance level.
the contributions to the VCM are invariant with respect to the
order of mechanisms and the period of experiments.
4. Results related to the WTP elicitations
The impact of the revelation of information on average WTP for
3 L of milk is shown in Fig. 3 for different subgroups and both prod-
ucts. These averages take into account all bids, including the non-
engaged bidders with WTP equal to zero. The WTP determined in
round E after the revelation of the indemnity will be studied at
the end of this section. The X-axis details rounds of choices j with
j = A, . . ., D, and the information preceding the choice leading to the
WTP elicitation. In Fig. 3, the indicators D isolate the significant
impact of a single round of additional information and/or between
the two types of milk for a same round. We test for the significance
of the WTP differences following a single round of information by
using the Wilcoxon test for matched paired samples and indicate
the significant differences at the 1%, 5%, and 10% levels.
Fig. 3 shows that, for a given round, WTP for a given product dif-
fer throughout the 3 charts, in particular for the initial WTP elicited
in the initial round A, for which instructions given by the organiz-
ers to participants were as neutral as possible. This comparison
between subgroups suggests an impact of the experimental design
on WTP for one product, as it will be detailed in the forthcoming
Tables 58.
Fig. 3 also shows that participants significantly react to messages
revealing the methane emissions and the omega 3 content with
health considerations. The effect of the second message on the exter-
nality is more uncertain regarding its statistical significance. Infor-
mation related to the methane and externality was not indicated
on bottles by a precise label.10 A shift in the WTP for a product follow-
ing messages on the methane and externality indicates the marginal
WTP for these attributes by participants becoming aware of these
specific attributes. Conversely, the omega-3 content was indicated
by a small label posted on the linseed milk (see note 2), which means
that WTP variations linked to the last message measures (a) the mar-
ginal WTP for participants who did not notice the label on omega 3 at
the beginning of the experiment and (b) the focalization on the
omega-3 impact for participants who already noticed the label on
omega 3 at the beginning of the experiment.
The successive messages lead to a statistically significant
decrease in the WTP for the conventional milk and a significant
increase in the WTP for the linseed milk.11 Except one case, the aver-
10
Four labels were posted on the GrandLait Candia bottle, the linseed milk: (1) the
label Filire Nutrition, Omga-3 Naturels (namely, Nutrition for a sector, Natural
Omega3) with two sentences mentioning an improved diet for cows leading to more
omega-3, (2) the label Saveur de lAnne 2012 (namely, Favor of the year, tested by a
panel of consumers), (3) AgriConfiance (namely, Agri-trust, our commitment, your
guarantee) and (4) La Route du Lait (namely, Road of Milk, Sustainable Development
Approach).
11
An analysis of the frequency of milk purchase declared in the exit questionnaire
reveals that the participants only decreasing their WTP have a higher frequency of
purchase compared to the ones only increasing their WTP. Second, the participants
only decreasing their WTP tend to reduce their voluntary contribution to the public
good as shown in the next section.
Round D z-value p-value
WTPDConv,i 1.819 0.069
WTPDLin,i 1.650 0.099
age WTP changes in absolute value are lower than 0.10 (with a rel-
ative variation compared to the initial WTP lower than 5%), which
suggest a significant but relatively small impact of each round of
additional information on WTP. These marginal WTP measure the
WTP linked to a characteristic, and the premium for the linseed
reducing the methane emissions seems a little more important than
the premium for the omega-3 that comes in last position.12
Interestingly, we find that the WTP for both types of milk in
April were influenced by the orders between the WTP elicitation
mechanism and the VCM, as visible in Fig. 3. This result is checked
with Table 5 focusing on the April experiment. When the VCM was
preceding the WTP elicitation, the WTP are significantly lower than
the WTP of participants for whom the WTP for the private good
were elicited before the VCM. In other words, Table 5 shows that
the WTP differ across both subgroups for which orders between
games differ. Importantly, these results differ from the VCM, in
which the order between both mechanisms did not influence the
contribution. Note that one way to limit WTP upward biases in
auctions would consist in organizing a VCM or some other mecha-
nism before the WTP elicitation.
With the variations in WTP coming from information between
two rounds for a same product, the reader can infer from a look
at Fig. 3 that the marginal WTP for attributes revealed in messages
are relatively similar throughout the three graphs. Some statistical
tests, not detailed in this draft, confirm this visual impression by
showing that the impact of information on WTP for attributes
given by the WTP differences between two rounds are very close
in average and not statistically different (1) across the subgroups
with different orders of mechanisms in April and (2) across both
experiments in April and November.
We now turn to the econometric estimation related to WTP.
Given that each participant i wrote 4 WTP, there should be some
correlations across data points relating to WTP. Furthermore,
WTP for a specific product cannot be negative and is left-
censored at zero, which explained that we use the random effects
Tobit estimator in the first two columns of Tables 6 and 7. As WTP
differences between products and variations in WTP coming from
information are estimated in the other columns on the right in
Tables 6 and 7, differences are not censored a priori, explaining
the use of a random effects estimator. In all columns of Tables 6
and 7, we tested for the influence on WTP of participants socio-
economic characteristics and their contribution to the VCM. We
tested for the influence on contribution of many variables and
re-run regressions without the variables characterized by a
12
Regarding the revelation of information on methane, 33.5% of participants are
either indifferent to the information with constant WTP or interested by the
information but mainly with dichotomous reactions. This dichotomy means that
participants mainly change the WTP of only one product, while the one of the
alternative product is constant. 20.1% of participants decrease their WTP for the milk
without linseed, while their WTP for the product with linseed is constant after the
revelation of information. Alternatively, 24.6% of participants increase their WTP for
the milk with linseed, while their WTP for the product without linseed is constant after
the revelation of information. Only 10.4% of participants simultaneously increase
their WTP for the milk with linseed and decrease their WTP for the product without
linseed.
 S. Marette et al. / Food Policy 66 (2017) 1224 21

Table 6
WTP elicited in April.

Dependent variable WTP WTP differa WTP variation/infob

Specification Rounds A, B, C, D Rou. A, B, C, D Rounds B-A, C-B, D-C
Tobit random effects Linear rand. Linear random effects
Model Convent. Linseed Linseed Conv. Convent. Linseed
Information on environment 0.118 0.106 0.221 0.116 0.105
Dummy = 1 if round B (0.037) (0.039) (0.034) (0.025) (0.021)
Information on externality and public impact 0.154 0.108 0.258
Dummy = 1 if Round C (0.037) (0.039) (0.034)
Information on omega 3 (0/1) 0.197 0.159 0.351 0.051
Dummy = 1 if round D (0.038) (0.039) (0.034) (0.021)
Order of mechanisms (April only) 0.353 0.366 0.045
Dummy = 1 if VCM WTP (0.038) (0.037) (0.022)
Contribution to choice #1 in VCM 0.137
(0.043)
Contribution to choice #2 in VCM 0.356
(0.021)
Agec 0.010 0.015
(0.001) (0.001)
Person in charge of purchasing the food, Dummy = 1 if the participant is in chargec 0.347
(0.041)
Number of kidsc 0.126
(0.018)
Monthly net income, 1 for income < 10007 for income > 6000c 0.037
(0.015)
Constant 1.936 1.551 0.305
(0.067) (0.064) (0.046)
Stand. dev. E 0.305 0.318
(0.010) (0.010)
Stand. dev. l 0.601 0.682
(0.014) (0.012)
Observations N = 536 N = 536 N = 536 N = 402 N = 402
Log likelihood 315.532 354.619
Adj. R2 0.168 0.046 0.052

Note: : significant at 1%; : significant at 5%; : significant at 10%. Standard errors in parentheses.
a
WTP difference between products, WTPXLin,iWTPXConv,I for a same round X = {A, B, C, D}.
b
For a given product, the variation of WTP between two successive rounds.
c
Question from the exit questionnaire.

Table 7
WTP elicited in November.

Dependent variable WTP WTP differa WTP variationb

Specification Rounds A, B, C, D Rou. A, B, C, D Rounds B-A, C-B, D-C
Tobit random effects Linear rand. Linear random effects
Model Convent. Linseed Linseed Conv. Convent. Linseed
Information on environment 0.152 0.097 0.237 0.142 0.075
Dummy = 1 if round B (0.043) (0.045) (0.037) (0.021) (0.018)
Information on externality and public impact Dummy = 1 if Round C 0.166 0.119 0.269
(0.043) (0.045) (0.037)
Information on omega 3 (0/1) 0.212 0.142 0.334 0.043
Dummy = 1 if round D (0.043) (0.045) (0.037) (0.021)
Contribution to choice #1 in VCM Dummy = 1 if contribution 0.998 0.040
(0.071) (0.015)
Patience, 1 for receiving 1000 euros in 1 year or 2 for receiving 1100 euros in 2 years c
0.485 0.604
(0.047) (0.079)
Number of kidsc 0.264
(0.048)
Monthly net income, 1 for income < 10007 for income > 6000c 0.091
(0.012)
Political opinion, Dummy = 1 if conservative, namely from the right (0 otherwise)c 0.360
(0.140)
Constant 1.095 0.780 0.312
(0.071) (0.122) (0.059)
Stand. dev. E 0.318 0.329
(0.011) (0.011)
Stand. dev. l 0.733 0.781
(0.015) (0.016)
Observations N = 452 N = 452 N = 452 N = 339 N = 339
Log likelihood 297.967 316.481
Adj. R2 0.171 0.055 0.092

Note: : significant at 1%; : significant at 5%; : significant at 10%. Standard errors in parentheses.
a
WTP difference between products, WTPXLin,iWTPXConv,I for a same round X = {A, B, C, D}.
b
For a given product, the variation of WTP between two successive rounds.
c
Question from the exit questionnaire.
22 S. Marette et al. / Food Policy 66 (2017) 1224

Table 8
WTP elicited in April and November, estimations with stacked data.

Dependent variable WTP WTP differa WTP variationb

Specification Rounds A, B, C, D Rou. A, B, C, D Rounds B-A, C-B, D-C
Tobit random effects Linear rand. Linear random
effects
Model Convent. Linseed Linseed Conv. Convent. Linseed
Information on environment 0.134 0.102 0.228 0.122 0.078
Dummy = 1 if round B (0.028) (0.031) (0.025) (0.018) (0.014)
Information on externality and public impact Dummy = 1 if Round C 0.160 0.112 0.263
(0.028) (0.031) (0.025)
Information on omega 3 (0/1) 0.204 0.151 0.343 0.036
Dummy = 1 if round D (0.028) (0.031) (0.025) (0.018)
Order of mechanisms, dummy = 1 if VCM WTP in April, 0 otherwise in April and November 0.246 0.288 0.046
(0.049) (0.048) (0.016)
Contribution to choice #1 in VCM 0.574 0.594
(0.035) (0.036)
Loss if overall contribution to VCM is lower than the threshold, dummy = 1 if loss in November 0.416
(0.046)
Period of the experiment, dummy = 1 if November 0.207 0.179 0.009 0.012 0.027
(0.059) (0.044) (0.067) (0.017) (0.019)
Person in charge of purchasing the food, Dummy = 1 if the participant is in charge c
0.089
(0.041)
Political opinion, Dummy = 1 if conservative, namely from the right (0 otherwise) c
0.128
(0.012)
Monthly net income, 1 for income < 10007 for income > 6000 c
0.177
(0.013)
Constant 1.512 1.489 0.304
(0.046) (0.070) (0.048)
Stand. dev. E 0.309 0.348
(0.008) (0.008)
Stand. dev. l 0.596 0.728
(0.013) (0.010)
Observations N = 988 N = 988 N = 988 N = 741 N = 741
Log likelihood 636.508 752.906
Adj. R2 0.169 0.075 0.036

Note: : significant at 1%; : significant at 5%; : significant at 10%. Standard errors in parentheses.
a
WTP difference between products, WTPXLin,iWTPXConv,I for a same round X = {A, B, C, D}.
b
For a given product, the variation of WTP between two successive rounds.
c
Question from the exit questionnaire.

coefficient that was not statistically significant. Estimations are
given in Table 6 for the April experiment and in Table 7 for the
November experiment (see the left parts of both tables).
The WTP for both types of milk significantly depend on the
revealed information as already underlined in Fig. 3. For the three
types of information, the coefficients for the conventional milk are
negative and the ones for milk with linseed are positive in Tables
6 and 7. The differences between coefficients of both products
for a given round of information underline positive WTP for the
public attributes linked to the global warming, and the private
attribute linked to omega-3.
For the April estimation in Table 6, the effect of the order of the
mechanism (with a dummy = 1 if the VCM precedes the WTP elic-
itation for milk) is significant with a negative coefficient which
confirms the negative impact on WTP. As shown in Fig. 3, starting
the experiment with another game significantly reduces the aver-
age WTP. Note that this significant effect is very close for both
milks, with coefficients equal to 0.353 and 0.366 in Table 6.
Some socio-demographic variables significantly influence WTP in
April and November. The age in April is the only socio-
demographic variable influencing the two types of milk in a similar
way.
In both tables, the contribution to the VCM #1 impacts the WTP
of some products. In April, this contribution significantly and pos-
itively impacts the WTP of the conventional milk, while it signifi-
cantly or positively impacts the WTP of the milk with linseed. This
influence is not systematic, since the WTP of the milk with linseed
in April and the WTP of the conventional milk in November are
not influenced by this VCM.
In the middle (third) column of Tables 6 and 7, the WTP differ-
ences between the two bottles of milk at a same round, namely
WTPXLin,iWTPXConv,I for a round X = {A, B, C, D}, are only influenced
by the revelation of information. The revelation of information reg-
ularly increases the WTP difference between two bottles, with the
highest coefficient related to the information regarding the private
attribute, namely the omega 3. The other variables linked to the
design of experiments are not reported because they were not sta-
tistically significant, which suggests a strong invariance of WTP
difference between products with respect to different contexts or
designs. All the coefficients of the third column of Table 6 in April
are very close to ones of Table 7 in November. This is a major dif-
ference with WTP for the two bottles of wine (see Table 1 p. 76 in
Ariely et al., 2003), for which WTP differences across different sub-
groups were very important.
The WTP variations after different rounds are detailed on the
right of Tables 6 and 7. There are some correlations across data
points since each participant is concerned by three differences.
WTP difference can be negative and positive, which explained that
we use the random effects linear estimator. WTP variations depend
on some dummy variables related to the rounds of information and
to the contribution to the VCM only for the WTP of the milk with
linseed. Once more, the link between both mechanisms is not
systematic.
In Table 8, data from April and November are combined, and we
added a dummy variable related to the Period of the experiment and
equal to one for observations of November (and zero in April). In all
the columns of Table 8, we kept this variable Period of the experi-
ment, even if not significant, for focalizing on the possible impact
 S. Marette et al. / Food Policy 66 (2017) 1224 23

Table 9
Impact of the revealed indemnity on the WTP.

April November
Number of sessions 11 8
Number of sessions with an increase of the indemnity linked to% contributors > threshold 10 4
% Participants who did no change their WTP 48.5% 68.1%
% Participants with at least 1 WTP decrease 23.1% 16.8%
% Participants with at least 1 WTP increase 23.1% 10.6%
% Participant with 1 WTP increase & 1 WTP decrease 5.2% 4.4%
Convent. Linseed Convent Linseed
Average change, E(WTPEWTPD) 0.011 0.013 0.04 0.08
Econometric estimation WTP variation
OLS Round var. E-D
Convent. Linseed
Surprised by the indemnity 0.153
Dummy = 1 if yes (0.066)
Satisfied with the indemnity 0.243 0.350
Dummy = 1 if satisfied (0.102) (0.151)
Constant 0.230 0.402
(0.103) (0.145)
Observations N = 113 N = 113
Adj. R2 0.11 0.05

Note: : significant at 1%; : significant at 5%; : significant at 10%. Standard errors in parentheses.

of the change in the period of the experiment. Table 8 clearly
shows that WTP are influenced by many variables detailed in
Tables 5 and 6 (1st and 2nd columns), including the order of mech-
anisms particularly studied in April. For each product, WTP are sig-
nificantly influenced by the period of the experiment, which
confirms the sensitivity of WTP to variations in mechanisms and
periods. Conversely, the WTP difference between products (3d col-
umn) and the variations of WTP coming from information (last col-
umns) are not influenced by the time.
Eventually, the WTP determined in round E after the revelation
of the indemnity are studied in Table 9, underlining change in WTP
between round D and round E for a significant percentage of partic-
ipants. It is hard to fully explain the impact of the indemnity vari-
ation on the change in WTP. Indeed, 48.5% of participants in April
and 68.1% of participants in June did not modify their WTP after
the revelation of the indemnity, which confirms the relative neu-
trality between both mechanisms. No variable is able to explain
changes in the April experiment, while a few variables significantly
explain the variation in November. In the simple econometric esti-
mation at the bottom of Table 9, the negative coefficients related to
the constant capture the fact that a little more participants are
decreasing their WTP compared to the ones increasing their
WTP. The satisfaction with the indemnity positively counterbal-
ances the reduction in WTP by some participants.
For summarizing, this section shows that the elicitation mech-
anism underlines a significant and positive WTP for the public
attribute related to the methane emissions. While the WTP for pri-
vate goods are particularly sensitive to the order of mechanisms,
the marginal WTP for the public attribute are invariant across this
order.
5. Conclusion
The combination of two experiments (or mechanisms) in one
session leads to interesting insights regarding the stability of
choices. The contributions to the VCM are invariant with respect
to the order of mechanisms and the period of experiments. Regard-
ing private goods, the WTP for each milk bottle are particularly
sensitive to the order of mechanisms and to the period of the
experiment. Conversely, the WTP difference between products
and the variations of WTP coming from information about public
or private attributes are invariant across the orders of mechanisms
and periods of experiments. The WTP variations coming from
information underline positive WTP for the public attributes linked
to the global warming and the private attribute linked to omega-3.
These results of stability confer validity to experiments for mea-
suring marginal WTP for different attributes related to food. It
means that we can use these variations of WTP for credible welfare
estimates and evaluations of policies ex ante, namely before the
complete implementation of regulations on markets. This is for
instance the case with Rousu et al. (2014) who introduce an
approach to valuing information and regulatory options, with a
surplus measure that couples variations in WTP coming from infor-
mation under a non-hypothetical experimental auctions, with
time-series revealed preference demand estimates and prices of
products observed in supermarkets at the time of the experiment.
This methodology directly integrating variations in WTP coming
from information can be used for estimating the impact of a new
product adoption or a quality improvement as well as for cost-
benefit analyses. In other words, as variations of WTP are similar
for samples of participants who represent the population of Dijon,
extrapolation of these results to the whole population of Dijon and
Burgundy is possible for a consistent welfare analysis.
WTP variations can be used for robust welfare measures of poli-
cies serving as both impetus and motivation for the related exper-
iments, which is the case with the numerous papers quoted in the
book written by Lusk and Shogren (2007). This is also the case with
experimental papers published in this journal, as the ones written
by Disdier et al. (2013), Hayes et al. (2002), Liaukonyte et al.
(2012), Marette and Millet (2014) and Stenger (2000), to quote a
few of them.
Despite flaws and biases coming from lab experiments, the
results of this study provide results regarding the validity of exper-
iments for measuring contributions to VCM and WTP for different
attributes related to food. We opened a path for new combinations
of mechanisms searching to understand the stability of preferences
for both public goods and private goods.
Acknowledgements
The Institut de la Recherche Agronomique (INRA) provided the
funding. We thank the editor and the reviewers for their com-
ments. All errors are ours.
24 S. Marette et al. / Food Policy 66 (2017) 1224
Appendix A. Messages
On separate sheets written in French, the precise messages are
translated from the original French.
A.1. The first message about the impact of linseed on the reduction of
methane emission before round B
Human activities emit more greenhouse gas emissions causing
global warming. French agriculture accounts for 20% in the emis-
sions of greenhouse gases. Among the greenhouse gases emitted
by agriculture, methane, mainly produced by ruminants including
cows via the eructation linked to their digestion is the most impor-
tant. The addition of linseed in the diet of cows significantly
reduces their production of methane.
Linseed is not included in the diet of cows that produced the
milk Carrefour. Linseed is included in the diet of cows that pro-
duced the milk GrandLait Candia.
A.2. The second message about the externality before round C
Purchasing the milk GrandLait Candia reduces greenhouse gas
emissions, which is beneficial to the environment and the citizens.
Purchasing the milk GrandLait Candia is beneficial to the citi-
zens, including the ones purchasing milk with a low-quality for
the environment, like the milk Carrefour.
A.3. The third message about the omega-3 content before round D
One liter of milk GrandLait Candia has 3 times more omega-3
than a liter of milk Carrefour. Omega-3 can effectively prevent car-
diovascular diseases. The presence of omega-3 in milk Candia
GrandLait comes from feeding cows with linseed.
References
Andersen, S., Harrison, G.W., Lau, M.I., Rutstrm, E.E., 2008. Lost in state space: are
preferences stable? Int. Econ. Rev. 49 (3), 10911112.
Alfnes, F., Rickertsen, K., Shogren, J.F., 2011. Unstable individual bids and stable
market demand. In: International Congress. European Association of
Agricultural Economists, Zurich, Switzerland. August 30September 2.
Ariely, D., Lowenstein, G., Prelec, D., 2003. Coherent arbitrariness: stable demand
curves without stable preferences. Quaterly J. Econ. 118 (1), 73105.
Becker, G.S., 1976. The Economic Approach to Human Behavior. University of
Chicago Press, Chicago.
Becker, G.M., DeGroot, M.H., Marschak, J., 1964. Measuring utility by a single-
response sequential method. Behav. Sci. 9 (3), 226232.
Bernasconi, M., Corazzini, L., Kube, S., Marechal, M., 2009. Two are better than one!
Individuals contributions to unpacked public goods. Econ. Lett. 104, 3133.
Blanc Bleu Coeur Association, 2012. Paris, France. <http://www.sante.gouv.fr/IMG/
pdf/bleu_blanc_coeur.pdf> (accessed 01.06.2016).
Bliem, M., Getzner, M., Rodiga-Lanig, P., 2012. Temporal stability of individual
preferences for river restoration in Austria using a choice experiment. J.
Environ. Manage. 103, 6573.
Brosig, J., Riechmann, T., Weimann, J., 2007. Selfish in the end? An investigation of
consistency and stability of individual behavior. MPRA Paper 2035, University
Library of Munich.
Brouwer, R., 2012. Constructed preference stability: a testretest. J. Environ. Econ.
Pol. 1 (1), 7084.
Cadsby, C.B., Maynes, E., 1999. Voluntary provision of threshold public goods with
continuous contributions: experimental evidence. J. Pub. Econ. 71, 5373.
Coats, J.C., Gronberg, T.J., Grosskopf, B., 2009. Simultaneous versus sequential public
good provision and the role of refundsan experimental study. J. Pub. Econ. 93,
326335.
Croson, R.T., Marks, M.B., 2000. Step returns in threshold public goods: a meta- and
experimental analysis. Exp. Econ. 2, 239259.
Corazzini, L., Cotton, C., Valbonesi, P., 2013. Too many charities? Insight from an
experiment with multiple public goods and contribution thresholds. Available
at SSRN: <http://ssrn.com/abstract=2307154> (accessed 01.06.2016).
Disdier, A.C., Marette, S., Millet, G., 2013. Are consumers concerned about palm oil?
Evidence from a lab experiment? Food Pol. 43, 180189.
Falk, A., Heckman, J.J., 2009. Lab experiments are a major source of knowledge in the
social sciences. Science 326 (5952), 535538.
Falk, A., Fischbacher, U., Gchter, S., Effects, S.I., 2013. Living in two neighborhoods
social interaction effects in the laboratory. Econ. Inq. 51, 563578.
Hayes, D.J., Fox, J.A., Shogren, J.F., 2002. Experts and activists: how information
affects the demand for food irradiation. Food Pol. 27, 185193.
Hasson, R., Lfgren, A., Visser, M., 2010. Climate change in a public goods game:
investment decision in mitigation versus adaptation. Ecol. Econ. 70 (2), 331
338.
Kagel, J.H., Roth, A.E., 2000. The dynamics of reorganization in matching markets: a
laboratory experiment motivated by a natural experiment. Quart. J. Econ. 115,
201235.
Kotchen, M.J., 2005. Impure public goods and the comparative statics of
environmentally friendly consumption. J. Environ. Econ. Manage. 49, 281300.
Levitt, S., List, J.A., 2007. What do laboratory experiments measuring social
preferences reveal about the real world? J. Econ. Perspect. 21 (2), 153174.
Liaukonyte, J., Rickard, B.J., Kaiser, H.M., Okrent, A.M., Richards, T.J., 2012. Economic
and health effects of fruit and vegetable advertising: evidence from lab
experiments. Food Pol. 37, 543553.
Liebe, U., Meyerhoff, J., Hartje, V., 2012. Testretest reliability of choice experiments
in environmental valuation. Environ. Resour. Econ. 53 (3), 389407.
Loomis, J.B., 1989. Test-retest reliability of the contingent valuation method: a
comparison of general population and visitor responses. Am. J. Agric. Econ. 71
(1), 7684.
Lusk, J.L., Schroeder, T.C., 2004. Are choice experiments incentive compatible: a test
with quality differentiated beef steaks. Am. J. Agric. Econ. 86 (2), 467482.
Lusk, J.L., Shogren, J.F., 2007. Experimental Auctions. Methods and Applications in
Economic and Marketing Research. Cambridge University Press, Cambridge, UK.
Marette, S., Millet, G., 2014. Economic benefits from promoting linseed in the diet of
dairy cows for reducing methane emissions and improving milk quality. Food
Pol. 46, 140149.
Martin, C., Rouel, J., Jouany, J.P., Doreau, M., Chilliard, Y., 2008. Methane output and
diet digestibility in response to feeding dairy cows crude linseed, extruded
linseed, or linseed oil. J. Anim. Sci. 86, 26422650.
McCarter, M.W., Savikhin, S.A., Sheremeta, R.M., 2013. Divided loyalties or
conditional cooperation? An Experimental Study of Contributions to Multiple
Public Goods (February 13, 2013). Available at SSRN: <http://ssrn.com/abstract=
2221696> (accessed 01.06.2016).
Moir, R., 2006. Multiple public goods and lottery fund raising. In: Isaac, R.M., Davis,
D. (Eds.), Experiments Investigating Fundraising and Charitable Contributors:
Research in Experimental Economics, vol. 11. JAI Press Inc, Stamford, CT, pp.
121142.
Nguyen, T.T.H., van der Werf, H.M.G., Eugne, M., Veysset, P., Devun, J., Chesneau, G.,
Doreau, M., 2012. Effects of type of ration and allocation methods on the
environmental impacts of beef-production systems. Livest. Sci. 145, 239251.
Roosen, J., Marette, S., 2011. Making the right choice based on experiments:
regulatory decisions for food and health. Eur. Rev. Agric. Econ. 38 (3), 361381.
Rousu, M.C., Marette, S., Thrasher, J.F., Lusk, J.L., 2014. The economic value to
smokers of graphic warning labels on cigarettes: evidence from combining
market and experimental auction data. J. Econ. Behav. Organ. 108, 123134.
Schaafsma, M., Brouwer, R., Liekens, I., De Nocker, L., 2014. Temporal stability of
preferences and willingness to pay for natural areas in choice experiments: a
test-retest. Resour. Energy Econ. 38, 243260.
Stenger, A., 2000. Experimental valuation of food safety: application to sewage
sludge. Food Pol. 25, 211218.
Taylor, L., Morrison, M., Boyle, K.J., 2010. Exchange rules and the incentive
compatibility of choice experiments. Environ. Resour. Econ. 47 (2), 197220.
Volk, S., Thni, C., Ruigrok, W., 2012. Temporal stability and psychological
foundations of cooperation preferences. J. Econ. Behav. Organ. 81 (2), 664676.
Wansink, B., Sonka, S., Hasler, C., 2004. Front-label health claims: when less is more.
Food Pol. 29, 659667.