2.1. Social versus Moral Dilemmas
In the European Union, the share of consumers buying sustainable product varieties is typically quite low [25
]. This also applies to the Netherlands, the country in which this study was implemented. The market share of sustainable products in the Netherlands usually does not exceed 5–10% [26
], yet surveys consistently show that Dutch consumers consider sustainability one of the most pressing societal issues (e.g., [29
]). Apparently, most Dutch consumers are not willing or able to voluntarily change their consumption patterns to increase sustainability.
To consider sustainability important without acting upon this belief is consistent with a social dilemma. A social dilemma arises if two conditions are met (for classic contributions on the role of social dilemmas in daily life, see for example Olson [5
] and Schelling [31
], and for an early theoretical analysis, see Sen [32
]): (i) collective welfare is maximized if all members of the community undertake a specific action; and (ii) individual members can maximize their private welfare by not
undertaking this specific action, but this will be at the expense of collective welfare. Indeed, the individual costs of buying sustainable product varieties are relatively high (with price premiums for such products amounting to 10–50% compared to conventional varieties), but society is best off if everyone
makes this individual contribution. However, individuals are better off free-riding on the investments made by others, than making investments in the collective good (sustainability) themselves. From this reasoning, it is not surprising that consumers support the idea of more far-reaching government measures to promote sustainable consumption—such measures would help to resolve the dilemma. In 2007, as many as 70% of Dutch citizens subscribed the proposition that the government should take the initiative in solving important societal issues [33
]. A recent example for this is the strong support of the Dutch population for stricter building regulations to improve their energy efficiency [34
However, the oftentimes fierce public debates about concrete measures suggest that also something else could be at play. Perhaps it is too easy for survey respondents to say that they consider sustainability important and that government measures are needed, and that this is the case not just in standard surveys (which pose questions on the respondent’s opinion, attitudes and behaviours) but also in stated-preference valuation techniques such as contingent valuation and discrete choice experiments. Due to the hypothetical nature of surveys, these studies may overlook the possibility that consumers find sustainable products simply too expensive. Indeed, most studies on preferences related to nonmarket environmental goods employ contingent valuation (CV) and discrete choice experiments (also known as conjoint analysis) techniques [35
]. While the literature often assumes that it is in the respondents’ own interest to truthfully answer a yes/no preference question (“Are you willing to pay X
euros for this project, yes or no?”), it appears that these survey techniques generally lead to an over
estimation of the true willingness to pay. In the economic literature, this phenomenon is known as hypothetical bias
, and in the social-psychological literature addressed as socially desirable answers
. A good overview of the nature and degree of hypothetical bias is provided by the meta-analyses of [36
], and by the literature reviews of Harrison [38
], and Harrison and Rutström [39
]. This bias arises because of “yea saying” (stating support for the socially desired project) or because of strategic considerations (when respondents say “yes” to a specific price even though they would not actually be willing to pay that price, just because the project is more likely to be implemented if they answer “yes” rather than “no”). Strategically or socially misrepresenting one’s preferences is cheap because the respondent knows that her answers will not have direct financial consequences, as the payment question itself is hypothetical. Clearly, the potential for such hypothetical bias is particularly important for the present research in which we aim to explain the difference between consumers’ (stated) support for policies promoting sustainable products, and their much less frequent purchase of such products. An additional or alternative explanation for this observed difference may be that people perceive sustainable consumption as a moral issue, so they take moral motives into consideration. When an individual is purchasing (food) products, he can be motivated by moral concerns as a citizen but also, and at the same time, care about product characteristics as a consumer [40
]. The phenomenon that individuals’ preferences and concerns do not always translate into purchase behaviour may be due to the dual entities that individuals may have, the so-called consumer-citizen duality, or discrepancy. Note that moral concerns may also include (restricting) freedom of choice of others.
2.2. A Field Experiment
To gain insight into whether sustainability issues are viewed as a social or a moral dilemma, we used both standard survey questions as well as an economic semi-field experiment (according to the classification by Harrison and List [42
], our study is not a natural field experiment as participants were aware of the fact that they took part in a research project). A key characteristic of economic experiments is that decisions have real
consequences—financial, or non-financial—for the decision maker herself, but maybe also for others. The participants in our experiment were a representative sample of Dutch consumers responsible for the food shopping in their household. Their households were provided with a budget (or “credit”) that they could spend on either the sustainable or the conventional variety of specific product group, and hence decisions had real (financial and sustainability) consequences rather than that decisions were just hypothetical. Compared to standard laboratory experiments (using students as subjects), an important and unique feature of our study is that participants made their decisions in the environment where household decision making with respect to groceries typically takes place—at home, and/or in the supermarket. This way, participants were confronted with the real, tangible consequences of their personal beliefs (and those of others) about abstract values such as “sustainability”—and hence participants are expected to be better able to picture the consequences of their beliefs and actions on abstract values such as sustainability [43
]. In addition, we believe that a distinctive and noteworthy aspect of our study is the large sample size: more than 600 household representatives participated in our experiment.
The experiment consisted of a number of stages; the general setup is displayed in Table 1
. Before conducting the experiment in its definitive form, we tested the design in qualitative and quantitative pilot studies. Based on the results of these pilot studies, the original design was further optimized [44
]. The aim of the qualitative pilot study was to check whether the information provided to the participants on conventional versus sustainable varieties of the product was clear and correctly understood. The quantitative pilot study was done to detect and solve any problems in the design before conducting the final experiment. After having carefully pretested the experimental design, we are confident that participants fully understood the instructions such that confusion can be ruled out as a factor driving our results. All information exchange took place through the Internet, and there was no contact between participants. In Phase 0, we cooperated with an online survey company (TNS-NIPO) to recruit households from their database that would be a representative sample of (non-vegetarian) households in the Netherlands. Our subject pool is a representative sample of the population of non-vegetarian households in the Netherlands because: (i) we started with a representative sample of Dutch households; (ii) we mentioned neither the topic nor the purpose of our study in our contacts with the respondents in the selection phase of the study; and (iii) 93% of the households who were invited for the initial survey, ended up participating in the experiment. For a general analysis of meat consumption in the Netherlands, see Gilsing et al. [45
] and also Dagevos and Voordouw [46
]. Respondents were asked whether they were willing to participate in a study, and, if so, whether they were available during the entire period in which the study was implemented, whether they ate meat, and who was responsible for the household’s groceries shopping. We also collected data such as disposable income, age, household composition and educational level.
The actual experiment took place in Phase 1, when participants had to decide whether or not to (jointly) address a sustainability problem. They were randomly divided into groups, and each participant was allocated an experimental budget (credit) to be spent on a product of which both a (less expensive) conventional and a (more expensive) sustainable variety were available in the supermarket. To minimize the potential impact of physical differences, the conventional and sustainable product varieties had to be as similar as possible. Furthermore, both varieties had to be widely available and known by most people. The product that we selected for our experiment, meat, meets these criteria. The physical properties such as quality and taste are more or less similar between conventional and organic meat [47
], and both varieties are widely available and known to most people. However, organic meat is on average about twice as expensive as non-organic meat. If objectively conventional meat and organic meat are very similar in all other respects, this does not mean that people also perceive them as very similar. This holds for the meat’s taste, its health consequences, and environmental impact (land use, energy intensity); see for example Gilsing et al. [45
]. Note that including or omitting this information is likely to affect the levels
of support for regulations, but not the treatment differences
. Similarly, although we cannot rule out the possibility that social desirability “pressure” may have affected the level of support, there is no reason to believe that the perceived social desirability pressure varies between treatments, such that treatment differences are unaffected. Households in a group vote as to whether the budgets they all received can only be spent on purchasing the sustainable product variety, organic meat, or whether every household is free to spend their budget on either organic or conventional meat (see Section 2.3
for details). The budget (credit) received was based on the size and composition of their household—seven euros for every adult in the household, and four euros for every child—and was sufficient to cover the additional costs of buying organic meat for one week. After the voting (still in Phase 1), we asked the participants a number of questions related to voting motives and their expectations regarding the voting behaviour of other people.
In Phase 2, participants were informed about the outcome of the votes in their group, and the implications for their household, in particular about which type of meat would be reimbursed. Afterwards households had one week to spend their budget on actual meat purchase(s).
After the purchases, period Phase 3 started in which we asked participants whether they were happy with the voting outcome in their group, and why they spent the money in a particular way. We also asked them to what extent they thought that the government should take stricter measures to promote sustainable production and consumption, and some other background questions.
Finally, in Phase 4, households submitted the receipts of their meat purchases. After verifying that their behaviour was in line with the group decision, their costs were reimbursed (up the maximum credit available for their household).
In the experiment, households were randomly allocated to six different treatments, which differed in either the rules dictating whether the budgets should be used to buy products of the sustainable variety, or in the number of households involved in the decision making process. Table 2
shows all the treatments as well as the number of participating households in each treatment. Each household participates in only one treatment (a between-subject design). Furthermore, all treatment differences are restricted to Phase 1, such that the procedures, information provided in all phases, etc. are the same across all treatments. Even though people may have different perceptions about potential differences between conventional and organic meat (see discussion in Section 2.2
), which may affect the share of participants buying organic meat in our six treatments, this should not affect any treatment differences
. We rely on randomization combined with sufficiently large numbers of participants in each treatment in order to control for the unobserved characteristics across groups.
We manipulate details of the experimental decision-making environment in four ways. (i) To examine if social dilemma considerations play a role, we vary the group size (from 1 via 31 to 61 households in a group). If social dilemma considerations are the most important factor driving behaviour, we expect the share of households voting in favour of the regulation to increase in the size of the group. In addition, we also explore (ii) whether voting outcomes differ if the voting outcomes are non-binding rather than binding; (iii) if it matters whether the decision problem is framed less restrictively (as a subsidy rather than as a binding regulation); and (iv) how people vote if households can conditionally commit themselves to a particular outcome, depending on the share of households willing to do the same. We will describe the main treatment in most detail, and focus on the difference between this treatment and the other ones when describing the other five treatments. The usage of a between-subjects design brings along the risk of responses being insensitive to variations in “the size of the public good” offered, as is well-documented in the stated preference valuation literature. Reasons why respondents’ valuations of a specific project (or policy) may not vary much with the amount of environmental benefits created include, among others, warm glow considerations and the embedding effect [50
]. However, using a within-subjects design (rather than between-subjects) may also result in biased outcomes because of, among other factors, anchoring and mental accounting bias [52
]. We are confident that if we find a lack of responsiveness of voting outcomes to, for example, variations in group sizes, this is not due to above the biases for two reasons. First, the information offered regarding group size is salient in our experiment because decisions have real (financial as well as animal welfare) consequences as opposed to hypothetical decision making in stated-preferences studies [54
]. Second, as stated before, our pre-tests indicated that the difference in sustainability outcomes (including animal welfare consequences) was perceived to be substantial for the different group sizes.
The main treatment, on which all other treatments were based, was the .so-called “binding ban” treatment. In this treatment, a group of a specific size, 31 participants in the basic variant, had to decide, by majority vote, whether to prohibit using the budget for purchasing the conventional variety of the product. The group’s decision (voting outcome) was binding. The participants were informed that if the average participating household in their group would decide to spend their budget on buying organic meat for one week, about one chicken would have a better life. However, if the group of 31 participating households decided to vote in favour, the extra costs per household of buying organic meat would be unchanged, while about 40 chickens would have a better life. The instructions also included information about the price difference between the conventional and sustainable variety of the product. The participants were asked to vote as to whether all members of the group should be obliged to use their (experimental) budget only for buying the sustainable alternative, or whether everyone should be free to spend it on the variety of their choice. The question was phrased as follows. The question posed was very similar in all treatments, but the exact formulation was adapted to details of the treatment (e.g., all 61 households when group size was 61). Note that we have not included words such as “prohibit” and “ban” in the experiment.
You are now asked to vote on the following proposition:
“All 31 households shall use their credit only for buying organic meat.”
□ I vote IN FAVOUR (I think that all 31 households in my group should be obliged to use their credit only for buying organic meat).
□ I vote AGAINST (I think that all 31 households in my group should be free to use their credit for buying organic or non-organic meat).
The vote was binding, and hence the weakly dominant optimal strategy for a participant is to vote truthfully and according to their (household) preferences [35
]. After the vote, all participants were informed about the voting outcome in their group (Phase 2). If the majority of the group had voted in favour, then the credit could indeed only
be used for buying organic meat. Hence, the purchases of meat during the week following the vote were reimbursed only if they complied with the majority decision.
The first treatment variation in the experimental design is the group size. We implemented the binding ban treatment with not just group sizes of 31 participating households, but also with 1 and 61 participants in a group (Treatments 1 and 3 in Table 2
). In these treatments, the text in the above-mentioned questions is adjusted as well as the information provided to the participants. In particular, we informed our participants in groups of one (61) household what the “animal welfare” consequences were of their whole group spending their budgets on organic as opposed to conventional meat: about one (80) chicken(s) would have a better life.
A second change involves the type of commitment (binding or advisory). In one treatment, the 31 participants in a group were asked the same question as to whether budgets should be spent on organic meat or not, but where the vote was only advisory (the non-binding ban, Treatment 4 in Table 2
). Thus, here, when voting, participants knew that the group decision was not-binding: all meat purchases would be reimbursed, regardless of the type of meat and the voting outcome.
Furthermore, we implemented a binding treatment in which the choice and consequences were similar to those in the binding ban treatment, but in which the situation was framed differently. In this so-called binding subsidy treatment (Treatment 5 in Table 2
) the 31 participants of a group could vote as to whether all budgets should be used to subsidize
the purchase of organic meat only of those who were willing to buy this type of meat rather than voting for or against a ban. If the group majority voted yes in the subsidy treatment, the price difference
between the organic and the conventional variety was reimbursed. As in the binding ban treatment, the group decision was binding.
Finally, we also implemented a treatment in which we used the so-called strategy method [57
] to ask participants as to whether they were willing to commit themselves to buying organic meat conditional on the number of other participants who were willing to do the same (the binding conditional choice treatment, Treatment 6 in Table 2
). Here the participants were asked to indicate for six different situations what their decisions were (see Table 4
). For example, one question was: “Are you willing to commit yourself to buying organic meat if 9–5 group members will also commit?” As in the main treatment, the outcome was binding.
Before presenting the results, we derive some predictions for expected differences between treatments. The treatments differed as to how (or by whom) it was decided where the budget could be spent on (the sustainable product variety, or the conventional one), but the rules regarding whether or not the credit was disbursed were always the same. If the outcome of the decision process in their group was that no households should spend their budget on the conventional product variety, they were reimbursed for the purchases of organic meat they made in the week in which the experiment took place. If the outcome of the decision process was that no regulations were to be imposed, all meat purchases (up to the allocated budget) were reimbursed independent of whether they purchased organic meat, or not. As organic meat is about twice as expensive as the conventional product variety and the budget per household was large enough for it to cover the extra costs of purchasing one week’s supply of organic meat, the budget was sufficient to cover the costs of one week of conventional meat or half a week organic meat.
Obviously, the binding ban treatment with just one participant is essentially an individual commitment treatment, where the participant decides to commit her household to buying organic meat, or not. If participants view buying organic meat as a social dilemma, the share of participants voting in favour of the ban should be increasing in group size with 1, 31 or 61 households (Prediction 1). After all, the extra costs of buying organic meat are unchanged but the environmental and/or social consequences of the ban are greater if the ban applies to more households. Similar considerations apply to the binding conditional treatment. Earlier experimental research has documented that a substantial fraction of people are conditional co-operators. For example, Fischbacher et al. [58
] report that in a public good experiment half of the people are willing to contribute more to a public good the more others contribute. In our binding conditional treatment, the vote is also a voluntary contribution decision, and if participants are conditional co-operators and view sustainability as a social dilemma, the share of participants willing to commit themselves to purchasing the sustainable product variety can be expected to increase in the share of participants willing to commit themselves (Prediction 2).
Regarding the comparison between the binding ban and non-binding ban treatments one would expect higher shares of people in favour of the proposition when the ban is not binding (Prediction 3). One reason is hypothetical bias—people’s tendency to signal their pro-sociality (yea-saying) if the outcome of a poll or question does not have any real consequences. Here one could always express one’s support for organic meat consumption in the referendum while deciding to buy conventional meat in the supermarket instead. Another reason is that people may view the non-binding vote as an opportunity to influence other people’s opinions and behaviour—expressing their support for organic meat in the referendum in the hope that it will induce other households to (also) buy organic meat.
A final comparison is between the binding ban and the subsidy treatment. These treatments are very similar, but the framing is slightly different. In the subsidy treatment households can vote whether the budget should be used to cover the extra costs of purchasing the sustainable product variety whereas in the main treatment they vote whether households should use their credit only for buying organic meat. From a cost–benefit analysis, voting in favour of the subsidy is (weakly) less attractive than voting in favour of the ban and hence, for given sustainability preferences, the participants are expected to be (weakly) less prone to voting in favour of the subsidy than to voting in favour of a ban (Prediction 4). The reason why voting in favour of the subsidy is weakly less attractive than voting for the ban is as follows. In the ban treatment, subjects receive compensation for their purchases of organic meat up to the maximum of the budget reserved for them. If they purchase the same amount of organic meat as they would normally do, they pay half of the bill themselves. In this case, the subsidy treatment and the binding ban treatment are identical. However, in the ban treatment, subjects can, in principle, just buy half of the normal quantity they typically consume per week, claim the entire budget, and buy additional meat of the conventional type. In that situation, voting in favour of the “policy” would be less costly to the participant in the binding ban treatment compared to the binding subsidy treatment since in the latter subjects would only receive half of the budget. The experimental results, however, indicate that this “strategy of just spending part of one’s budget on organic meat” (rather than 0, or all) is hardly ever used in the binding ban treatment.