1

Signaling Corporate Social Responsibility: Testing

Third-Party Certification vs. Self-Regulation in the Lab

Fabrice Etilé* and Sabrina Teyssier#

January 5, 2012

Abstract

Although consumer attitudes toward corporate social responsibility are positive, socially responsible products are far from gaining significant market shares. Information asymmetries have been identified as contributing to the attitude-behavior gap. Signaling may remedy this market failure. We use an experimental posted offer market to investigate the relative impact of certified labels, unregulated claims, and brands on sellers’ choice to supply socially responsible products and to signal it, and on consumers’ choice of ethical quality. When labels are certified by a third-party, or when consumers can identify ethical brands, a separating equilibrium emerges, whereby high and low-quality products are exchanged at different prices. However, efficiency gains are significant only under third-party certification. Unregulated claims are associated to ‘halo’ effects in consumer choices, which reduce their welfare as compared to a situation where signaling is not possible.

Keywords: labels, social responsibility, social preferences, separating equilibrium, posted offer market game. JEL codes: C92, D82, L15, M14.

* INRA – ALISS, UR 1303, 65, Boulevard de Brandebourg, 94205 Ivry-sur-Seine cedex, France, and Paris School of Economics, France. E-mail: Fabrice.Etile@ivry.inra.fr. # INRA – ALISS, UR 1303, 65, Boulevard de Brandebourg, 94205 Ivry-sur-Seine cedex, France. E-mail: Sabrina.Teyssier@ivry.inra.fr; corresponding author. Acknowledgments: We are grateful to Maxim Frolov for research assistance and Marie Font for administrative help. We gratefully acknowledge funding from the ANR ALIMINFO. We warmly thank participants at the Ecole Polytechnique seminar and at the conference on “The Economics of Corporate Social Responsibility” (Paris).

2

1. Introduction

Corporate Social Responsibility implies the use of social-, environmental- and health-

friendly technologies by companies.1 It has become one of the major priorities of managers in

the retail and consumer goods sector according to the Consumer Good Forum 2011 (REF:

Hartman, 2011). Opinion surveys also reveal that there is a growing interest of consumers in

CSR (Doane 2001).2 However, market shares remain quite low – less than 1% for most fair-

trade products in France.3 A key issue is the information asymmetry between sellers and

consumers, as the social responsibility incorporated in a product is a priori a credence

attribute (Darby and Karni 1973; Nelson 1974).4

Labels appear as a simple signaling device to mitigate this information asymmetry issue.

However, the CSR labels observed on actual consumer markets are heterogeneous with

respect to the level of commitment in socially responsible production processes that is

guaranteed. Some of them are delivered to firms after a certification process run by an

independent agency (a third-party), while others are merely marketing claims that are part of a

brand-building strategy. The newspaper The Economist suggested in an answer to the ‘No-

Logo’ movement that brands play the same signaling role as certified labels, because “they

make firms accountable to consumers”, and “brands of the future […] will also have to signal

something wholesome about the company behind the brand […] social responsibility”.5 This

paper asks whether efficiency in markets where CSR differentiates goods definitely requires

third-party certification, or whether it can be achieved softly by the combined effect of market

discipline, reputations and the awareness of media and NGOs, i.e. by self-regulation.

Our answer relies on empirical results from an experimental posted offer market, where

subjects are randomly assigned to the role of buyers or sellers and trade virtual goods. The

sellers have to choose a price, and the level of CSR that may differentiate in quality their own

1 The spectrum of activities covered by CSR is rather large, as social responsibility requires that attention be paid to many stakeholders beyond stock holders and consumers, in particular employees and suppliers, and be evaluated along dimensions such as human health, environment or local economic development. 2 In 2000, according to MORI, 66 per cent of European consumers declared that a CSR claim has triggered a purchase at least once C. Hines and A. Ames, Ethical Consumerism. A Research Study Conducted for the Co-Operative Bank by Mori (London: MORI, 2000). 3 French and U.S. consumers spent only 1.71 and 1.14 Euro respectively per year on purchases of fair-trade products in 2005, as against 19.02 Euros for the Swiss or 4.62 Euros for the British (Poret, 2007). 4 Such incorporation can be material as in aerosol products with no fluorocarbons, or just symbolic as in fair trade coffee. In general, consumers cannot assess the characteristics of the production process without enduring important information costs. 5 See The Economist, “The Case for Brands” (http://www.economist.com/node/771049/print) and “Who’s wearing the trousers?” (http://www.economist.com/node/770992/print), September 6th 2001, from the print edition.

3

offer. The precise quality remains private information. We test the impact of four

informational environments. In a baseline situation, there is no other signaling device than

the market price, and the buyers remain unaware of the exact quality throughout the game (No

Signaling treatment). In the research treatments, the sellers can signal the quality through a

label. We compare the case where the labels perfectly guarantee a Minimum Quality Standard

(MQS) certified by a third-party (the Third-Party treatment) to cases where labels are not

regulated by a third-party.6 They are claims that may be unsubstantiated. We examine two

mechanisms through which markets may discipline these unregulated CSR claims. In a first

treatment, there is a positive probability that unsubstantiated claims be detected and reported

to consumers. This simulates the pressure of NGOs and media on companies (Claim

treatment). A second treatment adds reputation effects to the Claim treatment, in order to

account for the relationships between brand identity and CSR (Brand treatment). We compare

these four treatments in terms of market efficiency and provision of social responsibility,

through the market shares of low- and high-quality products, their prices, and the sellers’

profits.

The experimental design proposes four key innovations in comparison with the existing

literature (REF). First, sellers do not only choose the prices and to label or not, but also the

level of quality that they want to offer. Second, product quality (social responsibility) is

measured by real donations to charities. These donations are chosen by sellers and increase

their production costs. Third, the virtual goods are pure credence goods, as the precise quality

of the offers is never revealed. They can only rely on prices, labels, and potentially detection

reports and the game history to infer the quality of offers. Last, we do not manipulate the

payoff functions to induce individual preferences over products quality. Letting individuals

express their actual preferences for social responsibility increases the external validity of the

experimental results. We use donations in a dictator game where receivers are charities to

control for the preference heterogeneity between the treatment groups.

Standard results from the theoretical research emphasize that imperfect information leads

to informational rents for sellers (REF). When signaling is not possible, buyers have no direct

means of learning about product quality. Yet, in the context of our experiment, they are aware

that some sellers may want to make donations and will offer high prices. Prices then become a

6 The word “label” was also used during the experiments, for the three research treatments, in order to avoid uncontrolled wording effects. It is clear however that each research treatment corresponds to a specific variety of signal, and is a way of representing the signalling content of certified labels, unsubstantiated claims and brands.

4

signal of quality, but those sellers who do not want to make donations have an incentive to

offer higher prices too: they will benefit from an informational rent, as all offers are pooled

together. Labels can reduce the informational rent and improve market efficiency only if

labeling is costly (REF). In this case, labels will convey truthful signal about products quality

and informational rents are reduced. A separating equilibrium will emerge, whereby those

consumers with strong preferences for social responsibility will buy high-quality, high-price,

products, and those who pay little attention to social responsibility will buy the standard, low-

price, products (Besley and Ghatak 2007; Conrad 2005; Rothschild and Stiglitz 1976; Spence

1973). We expect to observe such a separating equilibrium in the Third-Party treatment, as it

makes labeling costly. Efficiency gains should be obtained, because consumers are certain

that they make donations through the purchase of labeled goods; donations are increased and

informational rents are reduced. The outcomes of the Claim and Brand treatments are more

uncertain. On the one hand, labels do not reduce informational asymmetry, as all producers

can use them to claim that they reach the MQS. This is the usual ‘lemon’ problem (Akerlof

1970). On the other hand, being detected may produce immediate losses, and also future

losses if reputation is at stake. Hence, detection alone is perhaps not sufficient to overcome

the information asymmetry problem, but adding reputations may be sufficient and help to

achieve a separating equilibrium.

We indeed find a separating equilibrium in the Third-Party and Brand treatments: both

low- and high-quality goods are exchanged on the market at different prices. However,

efficiency is significantly higher in the Third-Party treatment, as sellers’ profits are lower and

the average quality of traded goods is higher, i.e. donations to NGOs are significantly higher

than in the Brand treatment. In the Claim treatment, labels are uninformative because almost

all products are labeled while production costs (and donations) are the same as in the No

signaling treatment. Yet, sellers make significantly higher profits as they are able to offer

higher prices. We interpret these additional profits without additional donations as an

indication that sellers exploit a ‘halo’ effect, whereby consumers are nudged to pay more for

labeled products even if labels are uninformative. In the end, third-party labeling with perfect

monitoring and a minimum quality standard does better than branding at promoting corporate

social responsibility and market efficiency.

The remainder of the paper is organized as follows. The experimental design and

procedures are presented in Section 2. Section 3 discusses the previous experimental evidence

on label regulations, and presents the main hypotheses that will be tested. Sections 4 to 6

analyze the results. Section 7 concludes.

5

2. Experimental design and procedures

The experimental design is structured in two parts. In each session, participants first play a

modified dictator game, in order to elicit their preferences for donations to NGOs. Then, they

participate in a posted offer market game.

2.1. Elicitation of preferences for donations

A modified version of the Dictator game is used to measure preferences for donations.

Each participant starts with an endowment of 50 ECU. Then, she has to decide how much she

wants to give to a receiver out of a list of four NGOs: ‘Emmaüs’, ‘the Red Cross’ and

‘Secours populaire’ are social NGOs helping poor or homeless people, and ‘Fonds ADIE’ is a

NGO that sustains entrepreneurship via micro-credit.7 Participants can give between 0 and 50

ECU, and keep the remaining sum. It is common knowledge that all decisions are anonymous,

and that NGOs will really receive the donations. While making their decisions in the Dictator

game, the participants do not know that they are going to participate in a market game.

2.2. A posted offer market game

After having played the Dictator game, subjects participate in a market game that is

repeated over 20 periods. The trading institution is a variation of the posted offer market (C.

A. Holt 2006; C. R. Plott and Smith 1978). At each market period, eight sellers and twelve

consumers trade a virtual good. The number of sellers is high enough to induce a price

competition in the absence of product differentiation.8 The roles are randomly determined at

the beginning of the market game and participants keep the same role during the whole game.

Participants trade using Experimental Currency Units (ECU), which are converted for

payments at the end of the experiment, at the rate of 8 ECUs = 1 Euro. The conversion rate is

known at the beginning of the experiment.

Four treatments were organized. Each subject participates in one treatment only. Whatever

the treatment, each period consists of four stages. First, sellers make their production

decisions and set up their offers. Second, offers are posted. Third, consumers make their

7 All these NGOs are well-known by the French population, except ‘Fonds ADIE’ but we provided details about the actions of each NGO at the beginning of each session. The diversity of the NGOs ensures that we can elicit preferences for donations unconditionally on the identity of the receiver. 8 It has been shown in the literature that competition on a market is almost perfect if there are four firms at least, as the entry of additional firms has then little effect on the market equilibrium (Dufwenberg and Gneezy, 2000; Krause et al., 2004).

6

decisions. Last, sellers and consumers get an informational feedback. We detail now the

baseline treatment. The next subsection will detail the main research treatments.

In the No Signaling treatment (NS treatment), each seller chooses a production cost, �, as

well as a price, �. The production cost is, at minimum, 20 ECUs per unit sold. The price must

be higher than or equal to the production cost. Sellers can choose a production cost higher

than 20 ECUs. Then, for every unit sold, the difference with the minimum cost �−20 is given

to a NGO, and sellers must choose which NGO will benefit from the donation among the

NGOs presented during the Dictator game. In this design, there are neither fixed costs of

production nor limited capacities of production: sellers produce the exact number of units

sold.9 All prices are then posted simultaneously and revealed to consumers. It is public

information that production costs over the minimum cost generate a donation to one of the

four NGOs, but consumers have no information about the production costs, the donations and

the NGOs that are chosen by sellers. The latter cannot be identified and tracked across

periods, as offers appear on the screen in a random order at each period. Consumers have a

fixed endowment, and must purchase at least four units of the virtual good.10 They can buy all

units to a single seller, or purchase units at different prices at several sellers. At the end of the

period, each consumer receives information about her own payoff (her endowment less her

expenditures); each seller is informed about her own profits, and the prices offered and

quantities sold by each seller. Then, a new period starts automatically. The payoff functions of

all agents (sellers, consumers and NGOs) are common knowledge. At each period, the payoff

�� of a seller � is computed as follows:

��=��+��−����,          ∀�∈1,2,…,8 (1)

where �� is the number of units sold by seller �, and a fixed payment of ��=50 ECUs is

added to guarantee a minimum gain to participants in the role of seller. The payoff of a

consumer � in every period is:

��=��−�=18�����,          ∀�∈1,2,…,12 (2)

where ��� is the number of units that consumer � purchased from seller � (note also that

�=112���=��), and ��=250 ECUs is a fixed endowment given at the beginning of the

9 We could assume limited production capacities for sellers as it is often assumed in industrial organization theory. However, this would raise methodological issues regarding the interpretation of the results. If consumers knew that production capacities were limited, they would be forced to make their choice more quickly. Some of them would then not have enough time to carefully choose their preferred option and this would generate noise. To obtain more reliable results, we therefore assume that sellers can always satisfy the demand. 10 As we do not want to formally induce consumer preferences through the payoff functions, as it is generally the case in posted offer market games, we force consumers to buy a minimal number of units (this framework is also used in Rode et al., 2008). This corresponds for instance to food choices or any goods that individuals have to purchase on a regular basis.

7

period. At the end of the period, donations made to an NGO � are:

��=�=18��−20�����,          ∀�∈1,2,3,4 (3)

where ��� is a dummy variable that equals 1 if seller � chooses NGO � as beneficiary, and

0 otherwise.

In this experimental design, the donation associated to each unit sold is not observed by

consumers: it is a credence attribute. From consumers’ point of view, products are

differentiated in quality according to the expected donation that each unit purchased can

generate. As emphasized in Section 2, we do not induce preferences over quality through

specific payoff functions for consumers. Their valuation of quality is their utility from making

donations, i.e. from purchasing goods whose production costs are expected to be higher than

20 ECUs. Consumer expectations regarding sellers’ production costs and donations depend on

their information set, which here includes only the prices posted by sellers. The other

treatments change the structure of this information set.

2.3.Main research treatments

The three other treatments have been designed in order to compare the effect of third-party

certification and market self-regulation. These two types of regulatory regimes differ by

requirements on labeling. Regulation by a third-party corresponds to official certifications as

ISO norms or labels administered by independent certifying agencies with strict criteria of

attribution such as Max Havelaar. Market self-regulation is not, strictly speaking, a regulatory

regime, as firms can use labels as a means of making unsubstantiated claim about their social

investments. Although there is no certification agency that controls whether the company has

any substantiation for its claims, media, activist groups or consumer associations can and

sometimes do assume this role. A key difference with third-party certification is that this

monitoring activity corresponds to a random probability that unsubstantiated claims be

detected.11 In addition, the consequences of detection differ according to whether its

reputation is at stake or not.

The Third-Party treatment (TP treatment) differ from the NS treatment in that sellers can

choose to post a label together with the price after having set their production cost. But they

can do it only if their production cost is higher than or equal to 25 ECUs – the minimum

quality standard (MQS) -, meaning that 5 ECUs at least will be given to an NGO. The label

thus indicates to consumers that a donation higher than or equal to 5 ECUs will be made with

11 Certification agencies also use random monitoring procedures, but only when the company has been certified for the first time. We here abstract from this consideration.

8

probability one if they purchase the good. This is common knowledge for all participants in

the session. The information set of consumers at the time of the purchase decision include the

two characteristics (price and label) of all posted offers. At the end of each period, they learn

their own payoff, while sellers can observe the characteristics of all offers and the quantities

sold by each seller.

In the Claim treatment (C treatment), sellers can also post their offers with a label but a

key difference with the TP treatment is that they can do it even if they choose a production

cost lower than 25 ECUs, which means that the label does not guarantee to consumers that a

donation of at least 5 ECUs will be made. Nevertheless, those sellers who post a labeled offer

but have chosen a production cost lower than 25 ECU have a one chance in three to be

detected as not fulfilling the MQS of 25 ECU. In this case, the offer appears on consumers’

screens with a warning message ‘Has the label but the production cost is lower than 25 ECU’.

The probability of detection is independently and identically distributed across sellers. In

comparison with the TP treatment, the information set of consumers at the time of the

purchase decision include an additional characteristic for all posted offers: the presence or the

absence of the warning message.

The Brand treatment (B treatment) is identical to the C treatment, except that sellers can be

tracked across periods. Each seller is identified by the same letter (from A to H) though the

whole game. The information set of consumers is potentially richer than in the C treatment, as

the identification of sellers may be used to link past and present offers.

2.4. Procedures

All sessions have been conducted at the University La Sorbonne in Paris. The design was

computerized with the software ‘Regate’ (Zeiliger 2000). The recruitment was made with the

software ‘ORSEE’ (Greiner 2004). We organized 22 sessions, with 20 participants in each

session: five sessions for the NS treatment, six for the TP treatment, six for the C treatment

and five for the B treatment.12

When participants entered in the experimental room, they were randomly affected to a

computer. They received instructions (see Appendix A) for playing the modified Dictator

game and details about actions of the four NGOs: ‘Emmaus’, ‘Red Cross’, ‘Secours

populaire’ and ‘Fonds ADIE’. The instructions were read aloud and then all participants

entered in the computer their decision without knowing that they were going to play a market

12 The average age was 26 and 48.7% of participants were men.

9

game with donations in the second part of the experiment. Once this game finished,

instructions for the posted offer market game were distributed. They were read aloud and

participants had to answer questions to check whether they understood the game rules

(payoffs functions, choices to be made, etc.). The game started only once all participants

correctly answered these questions. Participants were randomly assigned as eight sellers and

twelve consumers, and knew they were going to keep the role until the end of the game. All

treatments were framed in a neutral manner, and the participants knew that the donations were

going to be made for real.

Payments were made at the end of the session. One period out of the 20 periods of the

market game was randomly drawn to determine participants’ earnings in this game.

Participants received the sum of their earnings in the modified Dictator game, in the market

game, plus a show-up fee equal to 4€. Payments were received in a separated room to

preserve confidentiality. Each session lasted around 120 minutes and participants received on

average 23.20€. After the sessions, donations to NGOs were made online and proofs of

payments were sent back to participants.

3. Previous literature and research hypotheses

3.1.Experimental evidence on label regulations

In the presence of asymmetric information about products quality, labels are informative

only when labeling is costly (see for example). There are some experimental studies of

markets with adverse selection, wherein products are vertically differentiated and consumer

preferences for quality are homogenous. A robust empirical finding is that adverse selection is

reduced when a truthful signal about the quality of goods is introduced (Dejong et al. 1985;

Forsythe et al. 1999; C. Holt and Sherman 1990; Miller and Plott 1985; C. Plott and Wilde

1982). Cason and Gangadharan (2002) analyze in addition the effect of various label

regulations, as we do here. They use a posted offer market game, and compare the outcomes

of a Third-Party certification treatment, a Claim without detection or reputation treatment, and

a Claim without detection but with reputation treatment. Two key differences with our

experimental design are that products quality is revealed at the end of each market period in

every treatment and that they induce preferences for quality through payoff functions. Hence,

participants trade experience goods, not credence goods, and sellers have no preference for

quality while consumers have homogenous preferences. They focus on equilibrium selection

10

(high quality or low quality) rather than on the emergence of a separating equilibrium. They

find that third-party certification leads to higher efficiency gains while reputation alone is

insufficient to generate efficient outcomes.

In most experiments, preference for quality was implemented through a single consumer

payoff function. One exception is Rode et al. (2008), who conduct a laboratory market

experiment where three sellers and six consumers exchange units of a virtual good. The

sellers choose the prices but not the quality of products. The latter takes the form of a fixed

donation to an internationally recognized NGO fighting child-labor, and is associated to

higher production costs. Two sellers are committed to offer low quality goods (no donations is

made), while the last one offers high-quality products. The results reveal that high-quality is

offered at higher prices. When labeling or signaling is not possible, consumers purchase at the

lowest price: this is the ‘lemon’ equilibrium. When labeling is possible, many consumers

accept to reduce their monetary gains and to pay a premium for the labeled products: there is a

separating equilibrium.

We here combine and extend these studies. First, we endogeneize the choice of production

costs, product quality and labeling on the supply side. Second, we allow for heterogeneous

preferences over quality for both sellers and consumers. As higher production costs and

higher product quality are associated to donations made to NGOs, consumer will have

heterogeneous preferences over product quality because they differ in their preference for

donations. Introducing preference heterogeneity between sellers is equivalent to induce some

heterogeneity in production costs: the higher the preference for donations, the lower the

perceived cost of production. This experimental design is used to analyze the (co-)existence

of high quality and low quality goods on the market. Hence, we can analyze the existence of a

separating equilibrium under different regulatory requirements for labeling. Since participants

preferences are not constrained, the empirical results will arguably have a higher external

validity.

3.2.Research hypotheses

Although proposing a formal model for the experiment is far beyond the scope of this

paper, we now discuss in a non-mathematical language some predictions that can be made

using insights from the theory in Industrial Economics.

In our experiment, consumers have to purchase a product whose quality is uncertain. This

quality is the donation they make indirectly, through the seller’s production choice. They have

a preference for quality that is related to their preferences for donations, i.e. social

11

preferences. There is both vertical and horizontal differentiation: consumer utility increases

with quality, and for a given quality, utility is higher for individuals with strong social

preferences.

The consumers form their expectations about quality using the available information. Their

information set may include the label in the L, C, and B treatments, the warning message in

the C and B treatments, each sellers’ past offers in the B treatment, and potentially the current

prices and all past offers in all treatments.

Even in the No Signaling treatment, consumers may expect some sellers to have well-

developed social preferences, i.e., a taste for donations. Since sellers and consumers are

drawn from the same pool of individuals, we may suppose that it is common knowledge that

these sellers would like to choose higher production costs, in order to generate donations,

even if they cannot signal it through a label. For a given selling price, their subjective profit

(utility) increases when the production cost increases from 20 ECU, even if their monetary

profit decreases.13 Hence, they are likely to offer prices higher than 20 ECU. Nevertheless,

those sellers who do not have well-developed social preferences have some interest in

offering prices higher than 20 ECU as well. By doing so, they may expect to gain market

shares and make higher margins, since all consumers will have to pay higher prices. When

facing a set of offers at a price higher than 20 ECU, consumers know that only some of them

correspond to a positive donation. The price is therefore an imperfect signal for quality. Those

consumers with strong social preferences know that they take the risk of purchasing at a high

price a low-quality product. However, there is a strictly positive probability that a high price

offer corresponds to a high quality product, and they will still be willing to pay more than 20

ECUs. We should thus observe some transaction prices higher than 20 ECUs. As the

heterogeneity in social preferences generates heterogeneity in consumers’ valuation of quality

and in sellers’ preferences over production costs, we will observe a multiple price

equilibrium. The price distribution will depend on the distribution of social preferences in

participants, but also on competition between sellers. Since there are a limited number of

sellers and products are differentiated, there is a Bertrand competition, wherein sellers’

pricing strategies (conditional on their choice of production costs) depend on their

expectations about demand elasticity. Sellers are thus able to make positive profits for two

reasons. First, low- and high-quality sellers are pooled together in a unique price distribution.

13 Crampes and Hollander (1995) and Amacher et al. (2004) discuss formally the effect of heterogeneity in the cost structure of firms on labeling.

12

The former receive an informational rent, as they are able to sell low-cost product at a high-

price. Second, competition is not perfect.

Adding the certified label (the TP treatment) extends the range of the equilibrium price

distribution and contributes to separate the low-quality from the high-quality sellers. In this

case, sellers have to choose a production cost higher than or equal to 25 ECU to get the label.

Only those sellers who have strong social preferences will use both the label and the price to

signal their willingness to make a donation through the choice of a high production cost.

Those sellers with no social preferences will not find it profitable to choose a production cost

of 25 ECU in order to be pooled with high-quality sellers, as long as products at a price of 25

ECU or higher are not expected to represent a large market share. They will prefer to compete

for a larger market share, while maintaining their margin, by choosing a production cost

around 20 ECU and a price under 25 ECU.

Consumers observe whether goods are labeled or not knowing that the label is certified by

a third-party, i.e., that the label is truthful. This additional piece of information reduces the

asymmetry of information, since labeled products are necessarily of high quality. Those

consumers with strong social preferences will purchase labeled products at a high price, while

those consumers with no social preferences are likely to purchase unlabeled products at a low

price. The equilibrium in the L treatment is separating, such that two qualities are delivered

(Rothschild and Stiglitz 1976)(Spence 1973).

The price distribution will have two modes, one located between 20 and 25 ECU for

unlabeled products, and the other higher than 25 ECU for labeled products.14 The high-quality

products proposed by sellers with strong social preferences are separated from the other

offers. Hence, the average quality in a set of offers at a price under than 25 ECU is expected

to be lower as compared to the NS treatment, where high- and low-quality products are

pooled together. As a consequence, the average price of unlabelled products is likely to be

lower than the average price observed in the NS treatment.

The informational rent (and the profits) of low-quality sellers is largely reduced in this

treatment, but there are now two quality segments instead of a single one in the NS treatment,

which may affect price competition. The offers are less differentiated in each segment, but

there are less sellers per segment. The reduction in the number of sellers is likely to have little

impact for the segment of unlabelled products, as the number of competing sellers will

14 This prediction depends on the shape of the distribution of social preferences: we implicitly postulate here that it has only one mode.

13

probably be higher than four or five, which is enough to ensure perfect competition (see

Dufwenberg and Gneezy, 2000; Krause et al., 2004). Hence, the market for unlabelled

products will be more competitive than the market in the NS treatment. More competition and

less information asymmetry imply that the margins of unlabelled sellers will be lower than the

average margin observed in the NS treatment. There may be few sellers competing in the

market for labeled products. If this is the case, they may be able to make higher margins than

in the NS market. Although we do not have unambiguous predictions to propose, we expect

that the market will be more efficient: the average margins of all sellers should be lower, and

donations should be higher.15 This would imply a transfer of surplus from sellers to

consumers and NGOs, and does not exclude a transfer of profits from sellers with weak social

preferences to sellers with strong social preferences.

Following these arguments, the following hypotheses will be tested:

Hypothesis 1. In the TP treatment, both unlabeled and labeled goods are exchanged on the

market. There will be a price distribution with two modes, one under 25 and the other over

25, while the price distribution in the NS treatment will have only one mode, under 25.

Hypothesis 2. Efficiency is higher in the TP treatment than in the NS treatment:

consumers choose between labeled and unlabeled goods, sellers’ profits are unchanged or

decreased and donations are increased.

In the C treatment, sellers can post labeled offers even if their production cost is lower than

25. In this case, they face one chance out of three to be detected, but there is no reputation.

Detection is uninformative for all labeled offers with a price under than 25 ECU since the

production cost must be lower than the price. In this case, taking the label is riskless and thus

uninformative. Those sellers who want to offer a price lower than 25 ECU are indifferent

between taking or not the label. Only those sellers who would like to choose a price higher

than 25 ECU and a production cost lower than 25 ECU face a trade-off in their labeling

decision. There is the perspective of making higher margins, but there is the risk of being

detected and loosing current profits. If they take the risk, they will be pooled with high-quality

sellers, who also propose labeled products at a price higher than 25 ECU (but choose a

15 Efficiency gains are not perfectly observed, as they depend on the exact shape of social preferences. However, they are more likely when the agents can buy products whose quality better match their preferences (separating equilibrium), sellers’ profits are unchanged or reduced (there is more competition), and when donations to NGOs are increased (the production of public good increases).

14

production cost higher than 25 ECU). Both types of sellers will compete for consumers with

strong social preferences. Knowing that the offers are pooled, these consumers will be

reluctant to accept high prices, and the market share for the labeled products at a price higher

than 25 ECU will be quite small. This strategy will therefore not be profitable for those sellers

who would prefer to choose a production cost slightly under than 25 ECU. Most of them will

be better off at choosing a price lower than 25 ECU (and a slightly lower margin). It might be

worth taking a risk only for those sellers who have no social preferences and are willing to

choose low production costs. Then, offers at a price higher than 25 ECU are still likely to pool

high-quality with some low-quality products, and are therefore unlikely to attract a significant

market share. In the end, the information asymmetry is not reduced. The quality of offers is

uncertain as in the NS treatment, which generates some informational rent for those sellers

with no social preferences. We expect more offers to be labeled, but labels here do not reduce

the information asymmetry as they are costless (Matthews et al. 1991). Except the share of

labeled offers, the market outcomes should be very similar to those obtained in the NS

treatment: same price distribution, same margins, and same donations.

The Brand treatment adds reputation to the C treatment. Adding reputation changes a priori

nothing for offers at a price lower than 25 ECU: the label is costless and detection remains

uninformative. Sellers posting offers at a price lower than 25 ECU are indifferent between

choosing the label or not. As competition is not perfect, sellers may always make positive

profits in the future, and the cost of detection is much more important than in the C treatment.

Those sellers who take the risk of posting a labeled offer at a price higher than 25 ECU while

choosing a production cost lower than 25 ECU may loose their current and future profits.

They are less likely to take this risk, and most of them will choose a price under than 25 ECU.

They will be indifferent between having a label or not, which implies that less offers will be

labeled. Offers at a price higher than 25 ECU are much more likely to correspond to high

quality product, which restores the confidence of those consumers with strong social

preferences. The price will be a better signal of quality than in the NS treatment, and a better

signal of quality than the label. Although the label has no informative value per se, the

combination of labeling, detection and reputation separates out the high-quality products by

establishing brands. We may obtain a price distribution with two modes, one under 25 and the

other over 25, as in the TP treatment. Reputation restores the separation between low-quality

and high-quality sellers, and the margins and donations should be the same as those observed

in the TP treatment.

Hence, we have four more hypotheses to test:

15

Hypothesis 3. The market share of labeled goods is higher in the C treatment, than in the

TP treatment. The equilibrium prices in the C treatment are the same as in the NS

treatment.

Hypothesis 4. The market share of labeled goods is higher in the B treatment, than in the

TP treatment, and lower than in the C treatment. The equilibrium prices in the B treatment

are about the same as in the TP treatment.

Hypothesis 5. The equilibrium margins and donations in the C treatment should be the

same as in the NS treatment: efficiency is not increased.

Hypothesis 6. The equilibrium margins and donations in the B treatment should be the

same as in the TP treatment.

These six hypotheses will be tested with the experimental data. Before turning to the

results, it is worth emphasizing one important assumption behind the above economic

reasoning regarding the outcomes of the C and B treatments. It was assumed that all sellers

posting an offer with a price under than 25 ECU are indifferent between taking a label or not,

because in the case it is uninformative, as well as detection. But consumers may have a

preference per se for the label, and the detection message may also have a ‘moral’ value for

both sellers and consumers (i.e., “I do not want to be stigmatized”).

Section 5 tests Hypotheses 1, 3, and 4 by analyzing the market shares and prices. Section 6

focuses on efficiency and tests Hypotheses 2, 5 and 6.

4. Label regulations and the emergence of a separating equilibrium

This section analyzes the impact of label regulations on the market shares of labeled goods

and the prices. We compare results in the three treatments with label regulation to the No

Signaling treatment. As social preferences of sellers and consumers may drive their behavior,

we must guaranty that participants did not differ in terms of social preferences between the

four treatments. Participants’ social preferences are estimated via the dictator game played at

16

the beginning of the experiment.16 Kolmogorov-Smirnov tests comparing distributions of

donations between all treatments show no significant difference (at the 10% level) between

the distributions of social preferences across sessions, neither for sellers nor for consumers.

5.1. Third-party treatment

In the TP treatment, consumers can choose between labeled and unlabeled goods to the

extent that both types of product are offered by sellers. Contrary to previous experimental

work, consumers’ preferences are not induced via their payoff function. Hence, the resulting

demand indicates whether some consumers positively value the ethical quality of goods and

are willing to pay for it. Figure 1 presents the evolution of the share of labeled offers in , in

the TP treatment (the continuous line: the number of sellers offering labeled goods over the

total number of sellers) and the market share of labeled goods (the dotted line: the sum of

labeled goods bought by all consumers over the total number of goods exchanged).

Figure 1. Share of labeled offers and goods in the TP treatment

It shows that there is a positive demand for goods with a high ethical quality when labels

are certified by a third-party. On average, 31% (28.4% for periods 6 to 20) of exchanged

goods are labeled. This induces a positive externality on NGOs’ payoffs.

16 Indeed, we find that social preferences, i.e., subjects’ donations in the dictator game, are correlated to subjects’ behavior in the market game: linear regressions show that sellers’ donations in the dictator game are positively correlated to their production cost in the market game in the TP, C and B treatments and that consumers’ donations are positively correlated to the share of labeled goods they buy in the TP and B treatments. Hence, the Dictator game induces no ‘crowding out’.

0%  

20%  

40%  

60%  

80%  

100%  

1   2   3   4   5   6   7   8   9   10  11  12  13  14  15  16  17  18  19  20  

Periods  

%  labeled  offers  -­‐  L  Tr.  

%  labeled  goods  -­‐  L  Tr.  

17

Figure 2 presents the distribution of transaction prices observed in the market for labeled

and unlabeled products in the TP treatment, and all products in the NS treatment, for periods 6

to 20. The y-axis represents the market share associated to each transaction price. In the No

Signaling treatment, consumers mainly choose the minimal price or almost minimal price:

60.8% of goods are exchanged at price 21 and 20.0% at price 22. When Third-Party

certification is implemented, we observe a clear segmentation of the market. Unlabeled goods

are mainly exchanged at price 21 (84.5% of unlabeled goods) and labeled goods at price 26

(65.0% of labeled goods). Both unlabeled and labeled goods are sold at their (almost) minimal

price in the TP treatment.

Figure 1. Market shares depending on the price of goods in the NS and TP treatments

In addition, the market average prices, i.e., weighted averages of prices with weights equal

to the quantities purchased at each price, are stabilized after period 5 (see appendix B). For

the periods 6 to 20, the average price is 22.54 in the NS treatment. Labeled and unlabeled

products are exchanged at different prices in the TP treatment: the average price of unlabeled

goods is 21.31 and the average price of labeled goods is 27.39. The former is significantly

lower than the latter: the Wilcoxon signed-rank test statistics of the difference using the

average price for periods 6 to 20 is z=2.201, which is different from 0 at the level of 5% (p

=0.028). The average price of unlabeled goods is not significantly different from the average

price observed in the NS treatment: the Wilcoxon rank-sum test statistics of the difference

using the average price for periods 6 to 20 is z=1.333, which is not different from 0 at the

level of 10% (p=0.182). The average price of labeled goods is significantly higher than the

0%  

20%  

40%  

60%  

80%  

100%  

20   21   22   23   24   25   26   27   28   29   30   >30  

Price  

Market  share  -­‐  NS  Tr.  

Market  share  unlabeled  goods  -­‐  L  Tr.  

Market  share  labeled  goods  -­‐  L  Tr.  

18

average price in the NS treatment (Wilcoxon rank-sum test: z=2.000, p=0.046).

Hence, the experimental data confirm the existence of a separating equilibrium and

validates Hypothesis 1.

Result 1. In the Third-party treatment, both unlabeled and labeled goods are exchanged on

the market and these two types of goods are sold at different prices.

Figure 1 also reveals that the share of labeled offers is on average 48.1% (45.3% for

periods 6 to 20). Although the proportion of sellers offering labeled goods should adjust to the

demand, we observe that this is not the case. The difference between the share of labeled

offers and labeled goods is indeed significant: the Wilcoxon signed-rank test statistics using

the average market share for all periods is z=1.992, which is different from 0 at the level of

5% (p=0.046).17 Hence, sellers have obviously a non-strategic reason for offering labeled

goods. If this were not the case, the supply would progressively adjust to the demand.

Figure 2 indicates that prices of unlabeled and labeled goods in the TP treatment are very

close to the minimal possible prices of 20 and 25. This suggests that in this treatment

consumers do not rely much on prices to infer the ethical quality of offers. The observed

prices are very close to the equilibrium prices. In the NS treatment, 20.0% of goods are

exchanged at price 22, while offers at price 21 still exist. Hence, in the absence of signaling

the prices has some influence on consumers’ expectations regarding quality. This might create

an informational rent for some sellers.

5.2. Claim and Brand treatments

In the Claim and Brand treatments, sellers can receive a label although they choose a

production cost under 25. Figure 3 presents the evolution of the share of labeled offers in the

supply (the continuous lines) and the market share of labeled goods (the dotted lines) in these

treatments.

17 We obtain the same result if we only select periods 6 to 20, once behavior of subjects is more stabilized.

19

Figure 3. Share of labeled offers and labeled goods in the C and CR treatments

Figure 3 shows that 93.4% (92.7% for periods 6 to 20) of exchanged goods are labeled in

the C treatment. When reputation is introduced, this share reduces to 64.1% (64.9% for

periods 6 to 20). The reduction is statistically significant: the Wilcoxon rank-sum test

statistics of the difference using the average market share for all periods is z=2.008, which is

different from 0 at the level of 5% (p=0.045). Figure 4 presents the distribution of transaction

prices for each type of goods.

Figure 4. Market shares depending on the price of goods in the C and B treatments

0%  

20%  

40%  

60%  

80%  

100%  

1   2   3   4   5   6   7   8   9   10  11  12  13  14  15  16  17  18  19  20  

Periods  

%  labeled  offers  -­‐  C  Tr.  

%  labeled  goods  -­‐  C  Tr.  

%  labeled  offers  -­‐  CR  Tr.  

%  labeled  goods  -­‐  CR  Tr.  

0%  

20%  

40%  

60%  

80%  

100%  

20   21   22   23   24   25   26   27   28   29   30   >30  

Price  

Market  share  unlabeled  goods  -­‐  C  Tr.  

Market  share  labeled  goods  -­‐  C  Tr.  

Market  share  unlabeled  goods  -­‐  CR  Tr.  

Market  share  labeled  goods  -­‐  CR  Tr.  

20

In the C treatment, 46.0% of labeled goods are exchanged at price 21, 21.9% at price 22

and 30.3% at a price higher than or equal to 23. When reputation is introduced (the B

treatment), consumers mainly buy unlabeled goods at the minimal price: 86.6% of unlabeled

goods are exchanged at price 20. We observe that 54.1% of labeled-goods are exchanged at

price 21, 13.2% at price 22 and 31.4% at a price higher than or equal to 23. Hence, the

segmentation of the market in the B treatment is not similar to the segmentation observed in

the TP treatment, mainly because sellers have the possibility to post uninformative labels at a

price under than 25 ECU. In the B treatment, labeled goods are mainly sold at price 21 while

they are at price 26 in the TP treatment. More surprisingly, we observe much less labeled

offers at a price higher than 25 ECUs than in the TP treatment. This suggests that the

unconditional price distribution (i.e., not conditional on labeling) differs between the TP and

the B treatments.

In the C and B treatments, the market average prices are stabilized after period 5 (see

appendix C). Thus, Table 1 compares the average prices in all treatments for periods 6 to 20.

In the first column, we present the treatments, separating unlabeled from labeled goods. The

following columns indicate the p-value that corresponds to the Wilcoxon rank-sum test

statistics of a difference in prices between the treatments (one observation per session).

Average price Comparison to NS Tr. Comparison to TP Tr. Comparison to C Tr. Unlabeled goods

NS Tr. 22.50 - - - TP Tr. 21.31 p = 0.068 - - C Tr. 22.57 p = 0.855 p = 0.337 - B Tr. 20.24 p = 0.117 p = 0.465 p = 0.201

Labeled goods NS Tr.a 22.50 - - - TP Tr. 27.39 p = 0.006 - - C Tr. 23.53 p = 0.273 p = 0.025 - B Tr. 23.16 p = 0.251 p = 0.006 p = 0.361

All goods NS Tr.a 22.50 - - - TP Tr. 23.03 p = 0.273 - - C Tr. 23.46 p = 0.273 p = 1.000 - B Tr. 22.13 p = 0.601 p = 0.361 p = 0.584 a. In the NS treatment, we report the same average price for labeled goods and unlabeled goods although only

unlabeled goods are exchanged on the market. This is to simply present the significance of comparisons of average prices of labeled goods in the TP, C and B treatments and average prices of unlabeled goods in the NS treatment.

Table 2. Average prices in all treatments and significance of comparisons

The average prices of both labeled and unlabeled goods in the C and B treatments are not

21

significantly different from the prices observed in the NS treatment. The average price of

labeled goods is significantly higher in the TP treatment than in the B treatment. In the C

treatment, the average price of unlabeled goods is not significantly different from the average

price of the labeled good (Wilcoxon signed-rank test using the average price from period 6 to

20: z=0.105, p=0.917), while the average price of labeled goods in the B treatment is

significantly higher than the average price of unlabeled goods (Wilcoxon signed-rank test

using the average price from period 6 to 20: z=2.023, p=0.043). Hence, although branding

(labeling+detection+reputation) restores some form of separating equilibrium, it does not

yield the same price distribution as third-party certification. While Hypothesis 3 is not

rejected, Hypothesis 4 does not hold. This is our second result.

Result 2. In the C treatment, most offers are labeled and the price distribution is similar to

the distribution observed in the NS treatment. When reputation of sellers is introduced, i.e.,

in the B treatment, both labeled and unlabeled goods are offered at different prices, but the

price distribution is not the same as in the TP treatment.

Figure 4 indicates that prices of unlabeled and labeled goods in the C treatment and of

labeled goods in the B treatment are more dispersed than they were in the TP treatment. This

implies that consumers are more likely to use the prices as a signal of quality. However, there

are less offers at a price higher than 25 ECU in the B treatment than in the TP treatment,

which suggests that the price is a much less efficient signal of quality than a certified label.

One straightforward reason is that consumers are unable to develop the type of economic

argumentation that is proposed in Section 3.2. above. Branding is unlikely to have important

effects on sellers’ informational rent.

5. Label regulations and efficiency

Efficiency gains are observed when consumers can find offers whose quality match their

social preferences. In our experimental market, quality is uncertain, and efficiency can be

examined through the lenses of sellers’ margins and donations. Efficiency increases when the

margins are unchanged or reduced (an increase in sellers’ profits is synonym of an increase in

prices relatively to production costs), and donations to NGOs are increased. Tables 3 and 4

respectively present average sellers’ margins, i.e., the difference between sellers’ price and

sellers’ production cost, and average donations for both labeled and unlabeled goods per good

22

sold on the market. The p-value corresponding to the Wilcoxon rank-sum test comparing

average premiums and average donations between treatments are also presented. We consider

as independent observations the average premium and donation in each session from period 6

to 20 (the sum of payoffs and transfers of payoffs between agents, i.e., sellers, consumers,

NGOs, per period per treatment are reported in Appendix D).

Third-party certification leads to efficiency gains. First, the segmentation of the market

increases the average utility of consumers because labeled goods are unambiguously

associated to high-quality and consumers can make choices closely related to their social

preferences. Second, as shown in Table 3, the average margin on unlabeled goods is lower in

the TP treatment than in the NS treatment (although the difference is not significant), which

shows a reduction in sellers’ informational rent. The average margin remains high for labeled

goods, perhaps because segmentation reduces competition. We observe in Table 4 a

significant increase in donations from the NS treatment to the TP treatment (see the statistics

for all goods). In addition, the average donation is 5.22 for the labeled products, close to the

mandatory threshold of 5.

These results validate our Hypothesis 2.

Result 3. In the TP treatment, efficiency gains are observed as compared to the NS

treatment: consumers are provided with both unlabeled and labeled goods that correspond

to certified quality grades and better match their preferences; the average margins are

lower and the average donations are higher.

Average margin Comparison to NS Tr. Comparison to TP Tr. Comparison to C Tr.

Unlabeled goods NS Tr. 2.10 - - - TP Tr. 1.27 p = 0.273 - - C Tr. 2.29 p = 0.715 p = 0.337 - B Tr. 0.23 p = 0.175 p = 0.465 p = 0.144

Labeled goods NS Tr. 2.10 - - - TP Tr. 2.21 p = 0.855 - - C Tr. 2.94 p = 0.273 p = 0.200 - B Tr. 2.51 p = 0.347 p = 0.273 p = 0.465

All goods NS Tr. 2.10 - - - TP Tr. 1.53 p = 0.361 - - C Tr. 2.89 p = 0.361 p = 0.109 - B Tr. 1.71 p = 0.754 p = 0.855 p = 0.584

Table 3. Average premiums in all treatments and significance of comparisons

23

Average donation Comparison to NS Tr. Comparison to TP Tr. Comparison to C Tr.

Unlabeled goods NS Tr. 0.39 - - - TP Tr. 0.05 p = 0.006 - - C Tr. 0.27 p = 0.062 p = 0.733 - B Tr. 0.01 p = 0.009 p = 0.709 p = 0.842 Labeled goods NS Tr. 0.39 - - - TP Tr. 5.18 p = 0.006 - - C Tr. 0.59 p = 0.715 p = 0.004 - B Tr. 0.66 p = 0.016 p = 0.006 p = 0.100 All goods NS Tr. 0.39 - - - TP Tr. 1.50 p = 0.006 - - C Tr. 0.57 p = 0.855 p = 0.025 - B Tr. 0.43 p = 0.602 p = 0.011 p = 0.855 Table 4. Average donations per sold good in all treatments and significance of comparisons

In the C treatment, there are no efficiency gains as compared to the NS treatment. First,

Figure 3 shows that consumers can not use labels as a signal as almost all offers are labeled.

Second, Table 3 reveals that the average margin is slightly higher than in the NS treatment,

and significantly higher than in the TP treatment (see the lower panel for all goods). Third,

donations to NGOs are not higher in the C treatment than in the NS treatment (see Table 4).

In the B treatment, both unlabeled and labeled goods are offered on the market. Sellers’

margins are slightly lower than in the NS treatment, because the margin on unlabeled products

is close to zero. Perhaps surprisingly, the average margin is slightly higher than the margin

observed in the TP treatment. Although the average donation per labeled good sold on the

market is significantly higher than the average donation in the NS treatment, the average

donation per unlabeled good is significantly lower. On average, donations are not increased,

and are even significantly lower than in the TP treatment (0.42 as against 1.50 in the latter).

Therefore, there is some increase in efficiency, but it remains much lower lower than in the B

treatment. Hence, these empirical results rejects Hypothesis 6, and support Hypothesis 6.

Branding (the B treatment) leads to inferior outcomes than Third-Party certification, in terms

of sellers’ margins (they are higher) and in terms of donations (they are lower).

.

Result 3. In the C treatment, no efficiency gains are observed compared to the NS

treatment: only one type of goods is offered on the market, sellers make additional profits

and donations are not higher. Branding produces little efficiency gains as compared to

Third-Party certification.

24

Last, linear regressions are used to examine the differences in period payoffs of agents

between the treatments. We keep only period 6 to period 20, and the dependent variables are

the sellers’ profits, the consumers’ payoffs and the donations to NGOs, in the TP, C and B

treatments in comparison to the NS treatment. We control for a time trend. Standard errors are

adjusted for clusters in sellers when the dependant variable is sellers’ profits or donations to

NGOs and for clusters in consumers when the dependant variable is consumers’ payoffs.

Table 5 below presents the results. There are two significant results here. First, as expected,

donations to NGOs are significantly higher in the TP treatment. Second, sellers’ profits are

significantly higher in the C treatment, while they should not be different from 0. This result

cannot be explained by the informational rent from which sellers may benefit as the

information asymmetry is not stronger than in the NS treatment. Instead, we suggest that this

reflects a halo effect, which was originally defined by Edward Thorndike as “a problem that

arises in data collection when there is carry-over from one judgment to another” (Thorndike

1920).18 In the current experiment, the halo effect is a perception error of consumers about

quality that arises because the label is used as a signal of quality even when it is

uninformative. This signal tends to blur the other quality signal - the price -, perhaps because

it is easier to interpret. They tend to be overconfident in the label, because they are explained

that sellers have to choose high production costs to induce donations to NGOs through their

sales, and perhaps because they do not believe that sellers without strong social preferences

will try to cheat. In the NS treatment, consumers know that sellers may make donations to

NGOs through their sales but it is less prominent than in the C treatment as labels cannot be

used. Consumers seem to believe more in sellers’ social preferences when the ethical quality

of goods is emphasized although the way it is emphasized cannot be truthful. Therefore, an

imperfect label regulation without firms’ reputation should be avoided to preserve consumers’

well-being.

Result 4. In the C treatment, consumers are subject to a halo effect that decreases their

well-being.

18 E. L. Thorndike, 'A Constant Error in Psychological Ratings', Journal of Applied Psychology, 4 (1920), 469-77. was the first empirical study supporting the halo effect. The application was the rating of employees.

25

Dependant variable: Payoffs Sellers’ Consumers’ NGOs’ Periods -0.271***

(0.081) 0.400*** (0.079)

-0.152*** (0.056)

TP Treatment -3.960** -9.066** 7.505*** (1.734) (4.095) (1.411) C Treatment 4.505** -6.772* 1.309 (2.234) (3.735) (1.075) B Treatment -3.196* -2.838 0.553 (1.815) (2.984) (0.723) Donation in part 1 -0.106** -0.299** 0.135*** (0.054) (0.142) (0.037) Gender -0.406 -3.364 0.205 (1.386) (2.457) (1.026) Age -0.134* -0.643*** -0.022 (0.080) (0.241) (0.043) Constant 72.337*** 178.542*** 4.952*** (3.058)) (6.552) (1.608) Observations 2640 3960 2640 Prob > F 0.000 0.000 0.000 R-squared 0.056 0.122 0.079

Table 5. Linear regression explaining payoffs between treatments

However, as information asymmetries are lower in the TP and B treatments than in the NS

and C treatments, sellers’ informational rent should have been lower. Sellers’ profits are

indeed lower in the TP and CR treatments than in the NS treatment, but the difference is not

significant. This suggests that prices are somehow informative about the true product quality

when signaling is not possible, but branding and labeling do better. The informational rent

received by sellers in the NS treatment is reduced in the L and B treatments, because there is

less information asymmetry, but this reduction is partly offset by the decrease in price

competition due to the increasing differentiation of offers.

6. Conclusion

This experiment provides empirical results to compare the impact of certified labels,

unsubstantiated labels (i.e., claims), and branding on both the supply and the demand sides,

and on the ‘production’ of social responsibility. This was done without implementing the

consumers’ preferences for CSR through their payoff functions. Third-party certification

clearly leads to efficiency gains in comparison to a situation when signaling is not possible.

Branding leads to a separating equilibrium that is inferior to Third-Party certification, as

firms’ margins are higher and their donations are lower. Additionally, when unsubstantiated

quality claims about social responsibility are allowed, and firms’ reputation is not at stake, the

consumers support a halo effect that decreases their well-being and increases firms’ profits.

26

They misperceive the ethical quality of goods inferred from prices because the ethical quality

of goods is emphasized by the label and they do not believe that firms may cheat about it.

In terms of political recommendations, our data suggest that an imperfect label regulation

should be avoided for markets where reputation does not matter. The development of a market

for CSR products requires third-party certification, and can not be achieved through branding

strategies. Although certification fees must be supported, this is the only way to significantly

increase efficiency.

For further research, the results of this experiment suggest that consumers use the label as a

signal of goods ethical quality but they also use the prices. As suggested in the current paper –

and this is indeed an implicit assumption in Section 3.2., consumers believe that sellers have

social preferences as well. While this may be true when sellers and consumers are randomly

drawn from the same pool, it would be interested to have true company managers in the

position of sellers

References

Akerlof, G.A. (1970), 'The market for" lemons": Quality uncertainty and the market mechanism', The quarterly journal of economics, 84 (3), 488-500.

Amacher, Gregory S. , Koskela, Erkki , and Ollikainen, Markku (2004), 'Environmental Quality Competition and Eco-Labeling.', Journal of Environmental Economics and Management, 47, 284-306.

Besley, T. and Ghatak, M. (2007), 'Retailing Public Goods: The Economics of Corporate Social Responsibility.', Journal of Public Economics, 91, 1645-63.

Cason, TN and Gangadharan, L (2002), 'Environmental Labeling and Incomplete Consumer Information in Laboratory Markets', Journal of Environmental Economics and Management, 43 (1), 113-34.

Conrad, Klaus (2005), 'Price Competition and Product Differentiation When Consumers Care for the Environment.', Environmental and Resource Economics, 31, 1-19.

Crampes, Claude and Hollander, Abraham (1995), 'Duopoly and Quality Standards.', European Economic Review, 39 (1), 71-82.

Darby, M.R. and Karni, E. (1973), 'Free competition and the optimal amount of fraud', Journal of law and economics, 16 (1), 67-88.

Dejong, DV, Forsythe, R, and Lundholm, RJ (1985), 'Ripoffs, lemons, and reputation formation in agency relationships: A laboratory market study', Journal of Finance, 40 (3), 809-20.

Delpal, F. and Hatchuel, G. (2007), 'La consommation engagée s'affirme comme une tendance durable', CREDOC, Consommation et modes de vie, 201, 1-4.

Doane, D. (2001), Taking Flight: The Rapid Growth of Ethical Consumerism. (London: New Economics Fundation).

Dufwenberg, M. and Gneezy, U. (2000), 'Price competition and market concentration: an experimental study', International Journal of Industrial Organization, 18 (1), 7-22.

Forsythe, R, Lundholm, R, and Rietz, T (1999), 'Cheap talk, fraud, and adverse selection in financial markets: Some experimental evidence', Review of Financial Studies, 12 (3), 481.

27

Greiner, B. (2004), 'The Online Recruitment System ORSEE 2.0–A Guide for the Organization of Experiments in Economics', Papers on Strategic Interaction, University of Cologne, Working Paper Series in Economics, 10.

Hines, C. and Ames, A. (2000), Ethical Consumerism. A Research Study Conducted for the Co-Operative Bank by MORI (London: MORI).

Hirshleifer, J. and Riley, J.G. (1979), 'The Analytics of Uncertainty and Information-An Expository Survey', Journal of Economic Literature, 17 (4), 1375-421.

Holt, C and Sherman, R (1990), 'Advertising and product quality in posted-offer experiments', Economic Inquiry, 28 (1), 39-56.

Holt, C. A. (2006), Markets, Games, and Strategic Behavior (Boston: Pearson/Addison-Wesley).

Krause, M, Kroger, S, and Potters, J (2004), 'Insights from experimental economics for market regulation', Tijdschrift voor Economie en Management, 49 (2), 217-38.

Matthews, S., Okuno-Fujiwara, M., and Postelwaite, A. (1991), 'Refining cheap-talk equilibria', Journal of Economic Theory, 55, 247-73.

Miller, RM and Plott, CR (1985), 'Product quality signaling in experimental markets', Econometrica: Journal of the Econometric Society, 53 (4), 837-72.

Nelson, P. (1974), 'Advertising as information', The Journal of Political Economy, 81, 729-54.

Plott, C. and Wilde, L. (1982), 'Professional diagnosis versus self-diagnosis: An experimental examination of some special features of markets with uncertainty', in Vernon Smith (ed.), Research in Experimental Economics (2; Greenwich, CT: JAI Press), 63-112.

Plott, Charles R. and Smith, Vernon L. (1978), 'An Experimental Examination of Two Exchange Institutions', Review of Economic Studies, 45 (1).

Poret, S. (2007), 'Les défis du commerce équitable dans l'hémisphère Nord', Economie Rurale, 302, 56-70.

Rode, Julian, Hogarth, Robin M., and Le Menestrel, Marc (2008), 'Ethical Differentiation and Market Behavior: An Experimental Approach.', Journal of Economic Behavior and Organization, 66, 265-80.

Rothschild, M. and Stiglitz, J. (1976), 'Equilibrium in competitive insurance markets: An essay on the economics of imperfect information', The quarterly journal of economics, 90 (4), 629-49.

Spence, M. (1973), 'Job market signaling', The quarterly journal of economics, 87 (3), 355-74.

Thorndike, E. L. (1920), 'A constant error in psychological ratings', Journal of Applied Psychology, 4, 469-77.

Valor, C. (2008), 'Can consumers buy responsibly? Analysis of solutions for market failures ', Journal of Consumer Policy, 31 (3), 315-26.

Vermeir, I. and Verbeke, W. (2006), 'Sustainable food Consumption: Exploring the consumer "Attitute - Behavioral Intention" gap', Journal of Agricultural & Environmental Ethics, 19 (2), 169-94.

Zeiliger, R. (2000), 'A Presentation of Regate, Internet Based Software for Experimental Economics', http://www.gate.cnrs.fr/~ zeiliger/regate/RegateIntro.ppt, GATE, 2000.

Appendix

Appendix A - Instructions

28

Instructions (translated from French)

You are participating in an experiment on decision making for the Research Department ALISS-INRA and the Paris School of Economics. During this session, you can earn money. The amount of your gains depends on your decisions and on decisions of other participants with whom you have interacted. During the session, your gains are calculated in ECU, with the following conversion rule:

8 ECU = 1 Euro At the end of the session, your gains are determined and converted into Euros.

Additionally, you will receive a show-up fee of 4 Euros. Your gains will be paid in cash in a separated room to guaranty confidentiality.

You will participate in two different games in this session. Each game is totally independent of the other game: your decisions in one game will not influence your gains in the other game. Moreover, you will make decisions without knowing other participants’ decisions in the previous game. Note that other participants will not know your decisions either.

You already have instructions for game 1. Instructions for game 2 will be distributed once all participants will have made their decision in game 1.

To determine your final gains, your gain in game 1 will be added to the 4 Euros show-up fee and to the gain you will make in game 2. At the end of the session, your gains will be announced to you in private. The calculation of gains in each game is explained when we describe in details each game.

In the two games, all decisions are anonymous. If you have questions regarding the instructions, please raise your hand. We will answer

your questions in private. During the whole session, it is forbidden to communicate between each other.

Thanks for your participation. ___________

Game 1 -­‐ All participants receive 50 ECU. -­‐ Each participant decides which amount, between 0 ECU and 50 ECU, he wants to give to

an NGO among the four NGOs listed on the screen. o Croix Rouge o Emmaüs o Fonds ADIE o Secours Populaire

At the end of these instructions, you can find some information about the actions of these four NGOs. This information is from the website of each NGO.

-­‐ Each participant writes the amount (from 0 ECU to 50 ECU) he wants to give to an NGO. If the amount given is higher than 0 ECU, the participant selects the NGO to whom he wants to give this amount. Only one NGO is selected.

-­‐ Each participant validates his choice by clicking on the “OK” button.

-­‐ Payoffs: o Each participant receives at the end of the experiment: 50 ECU – the amount of his

donation to one of the NGOs. o Each NGO will effectively receive the sum of donations of all participants who

have chosen this NGO. Each NGO will receive money via a bank transfer done by internet after the session. In the next days all participants will receive via email the justification of bank transfers done for each NGO as well as their amount.

29

Example of calculation of payoffs: If a participant chooses to give 15 ECU to the Croix Rouge, the gain of the participant is

35 ECU. The Croix Rouge receives 15 ECU. If a participant chooses to give 30 ECU to Fonds ADIE, the gain of the participant is 20

ECU. The Fonds ADIE receives 30 ECU. ___________

Game 2 This game is independent of the previous game. This game consists of 20 periods. One period will be randomly chosen at the end of the

experiment to determine the amount of your gains. Your gains in the period that has been selected will be your final gains for game 2.

In this game, you will participate in a market. The market consists of height sellers and twelve buyers. You will receive one of these two roles at the beginning of the game. You will keep the same role during the whole game.

Sellers and buyers will exchange a virtual good on this market. All your decisions are anonymous.

Description of each period Each period consists in three stages:

In stage 1, sellers make their decisions. In stage 2, buyers receive information on sellers’ decisions in stage 1. In stage 3, buyers make their decisions.

Description of stage 1: Sellers’ decisions -­‐ All sellers receive an endowment equal to 50 ECU. -­‐ Each seller sells a virtual good on the market. He sells exactly the number of virtual goods

that are demanded to him. -­‐ Each seller must support a cost for each virtual good he sells. This unitary cost is called

“production cost”. The minimal production cost that each seller must support for each sold good is equal to 20 ECU.

-­‐ Each seller can make a donation to an NGO per sold good: Croix Rouge, Emmaüs, Fonds ADIE or Secours Populaire. Each seller chooses his production cost knowing that the NGO he chose will receive from him: (production cost – 20 ECU) × number of units sold. The production cost is necessarily higher than or equal to 20 ECU.

-­‐ Each seller chooses his production cost. -­‐ Each seller can obtain a “label” that will be observed by buyers. To receive this label, he

must tick the box “I wish to receive the label”. o If the seller does not tick the box “I wish to receive the label”, then the seller does not

receive the label. o If the seller has chosen a production cost at least equal to 25 ECU, i.e., if the seller has

made a donation to one of the four NGOs higher than or equal to 5 ECU, and if he has ticked the box “I wish to receive the label”, then the sellers receives the label. The seller is announced to buyers as having the label.

o If the seller has chosen a production cost between 20 ECU and 24 ECU (20, 21, 22, 23 or 24 ECU), i.e., if the seller has made a donation to one of the four NGOs equal to 0, 1, 2, 3 or 4 ECU, the seller cannot receive the label. The seller is announced to buyers as not having the label.

-­‐ The label does not indicate the exact amount of the seller’s production cost. -­‐ Each seller chooses the unitary selling price of his goods. The unitary selling price must

be between the production cost he has chosen and 62 ECU. -­‐ Each seller confirms his choices by clicking on the “OK” button. -­‐ While sellers make their decisions, buyers have a screen telling them to wait.

30

To summarize: o Each seller receives 50 ECU. o Each seller chooses a unitary production cost. This production cost must be higher

than or equal to 20 ECU. o If the seller has chosen a production cost higher than 20 ECU, he chooses to which

NGO he makes a donation. For each sold good by this seller, the NGO will receive: his production cost – 20 ECU.

o The seller chooses to receive the label or not if he has chosen a production cost higher than or equal to 25 ECU. If the seller has chosen to receive the label, the seller is announced to buyers as having the label.

o Each seller chooses the unitary selling price of the good. This selling price is between the chosen production cost and 62 ECU.

Sellers’ payoffs: Each seller makes a gain equal to:

50 + [Selling price – Production cost] × Number of sold units Donations to NGOs: Each NGO will receive for each seller who has chosen this NGO the following amount:

[Production cost – 20 ECU] × Number of sold units

In total, each NGO will receive the sum of the amounts that are given by sellers via a bank transfer done with internet at the end of the session. In the newt days, you will receive by email the justification of the bank transfers done to each NGO, as well as their amounts.

Information: At the end of the period, sellers receive information on: o Their own gains. o The decisions of each seller (without identifying them):

Their selling price. If they have the label or do not have the label. Their quantities of sold goods.

Description of stage 2: Information given to buyers

-­‐ All sellers’ offers appear on the screen of buyers. -­‐ The order of sellers’ offers is randomly chosen by the software at each period. The sellers’

offers are then never in the same order. The first offer on the screen at period 1 is from a different seller than the first offer on the screen at period 2. The first offer on the screen at period 2 is from a different seller than the first offer on the screen at period 3, etc. The sellers cannot be identified and the order of offers is random at each period.

-­‐ For each seller, buyers observe the unitary selling price and if he has the label (“has the label” or “does not have the label”).

Description of stage 3: Buyers’ decisions

-­‐ All buyers receive an endowment equal to 250 ECU. -­‐ Buyers must buy at least 4 goods. -­‐ Buyers decide to whom they buy their goods. No constraint is imposed on their purchases,

except that they must buy at least 4 goods and that they must not spend an amount higher

31

than 250 ECU. The software prevents from validating in this case. An error message will indicate to buyers that their decisions must be modified to be validated. The buyers can buy more than 4 goods if they wish. They can buy all their goods to the same seller. They can also buy goods to several buyers.

-­‐ As the buyer fills in the boxes corresponding to the bought goods for each offer, the total price to pay for each offer is directly calculated by the software. At the bottom of the screen, the “total number of bought goods” is also calculated, as well as the “total price of goods” which corresponds to the total amount paid by the buyer to all the sellers: this is the amount that the buyer will pay in total.

To summarize: o Each buyer receives 250 ECU. o Each buyer chooses the seller(s) to whom he buys his goods and the quantity of goods

he buys. He must buy at least 4 goods on the market, for a total price of goods lower than or equal to 250 ECU.

Buyers’ payoffs: Each buyer makes a gain equal to:

250 – total price of goods Information: At the end of each period, buyers receive information on: o Their own gains.

At the end of each period, a new period starts automatically. It is impossible to identify sellers’ offers between two different periods. The order of

sellers’ offers is randomly changed at each period. All decisions are anonymous. Example of calculation of gains Example 1

A seller has chosen a production cost equal to 22 ECU and a price equal to 40 ECU. Then, this seller cannot receive the label. This seller has chosen the NGO Secours Populaire. A buyer buys 4 goods to this seller. Gain of this seller = 50 + (40 - 22) × 4 = 122 ECU Gain of this buyer = 250 - (4 × 40) = 90 ECU Amount of the donation made by this seller to the Secours Populaire = (22 - 20) × 4 = 8 ECU

Example 2 A seller has chosen a production cost equal to 28 ECU and a price equal to 40 ECU. Then, this seller receives the label if he has ticked the box “I wish to receive the label”. This seller has chosen the NGO Fonds ADIE. A buyer buys 3 goods to this seller and 1 good to another seller on the market at a price equal to 21 ECU (this other seller has chosen a production cost equal to 20 ECU). Gain of this seller = 50 + (40 - 28) × 3 = 86 ECU Gain of this buyer = 250 - (3 × 40 + 21) = 109 ECU Amount of the donation made by this seller to the Fonds ADIE = (28 - 20) × 3 = 24 ECU

___________ Please, read these instructions carefully. Before starting the game, we will ask you some

questions on these instructions. Once you will have correctly answered these questions, the

32

game will start. ___________

Game 2 – Questionnaire on the instructions Please answer the following questions relative to these instructions.

You keep the role of the buyer (or of the seller) during the 20 periods of the game.

If you are a seller, your production cost is necessarily equal to 20 ECU.

If you are a seller, you choose the unitary selling price of the virtual good.

If you are a seller, you can receive the label only if your production cost is higher than or equal to 25 ECU.

If you are a buyer, you must buy 4 goods.

If you are a buyer, when you see a seller’s offer with a label means that the seller made a donation to an NGO at least equal to 5 ECU.

The order of sellers’ offers is modified at each period.

Right

□ □ □ □ □ □ □

Wrong

□ □ □ □ □ □ □

Example 1

A seller has chosen a production cost equal to 20 ECU and a price equal to 50 ECU. This seller has chosen the NGO Emmaüs. A buyer buys 4 goods to this seller. What is the gain of this seller? _______________ Can this seller receive the label? _______________ What is the gain of this buyer? ______________ What is the amount of the donation to Emmaüs? ________________

Example 2

A seller (first seller) has chosen a production cost equal to 30 ECU and a price equal to 45 ECU. This seller has chosen the NGO Fonds ADIE. Another seller (second seller) has chosen a production cost equal to 23 ECU and a price equal to 25 ECU. This seller has chosen the NGO Croix Rouge. A buyer (first buyer) buys 2 goods to the first seller and 2 goods to the second seller. Another buyer (second buyer) buys 1 good to the first seller and 5 goods to the second seller. What is the gain of the first seller? _______________ Can the first seller receive the label? _______________ What is the gain of the second seller? _______________ Can the second seller receive the label? _______________ What is the gain of the first buyer? _______________ What is the gain of the second buyer? _______________ What is the amount of the donation to the Fonds ADIE? ______________ What is the amount of the donation to the Croix Rouge? ________________

Appendix B – Evolution of prices in the NS and L treatments The figure below compares the average prices of labeled and unlabeled goods in the L

treatment to the average price of goods sold in the NS treatment. These prices are weighted averages with weights equal to the quantities purchased at each price.

33

Appendix C – Evolution of prices in the C and CR treatments The figure below compares the average prices of labeled and unlabeled goods in the C

treatment to the average price of goods sold in the CR treatment. These prices are weighted averages with weights equal to the quantities purchased at each price.

Appendix D – Sum of payoffs and transfers of payoffs per period per treatment The market game is a zero-sum game in the sense that the total amount of wealth is

identical in all sessions in all treatments. Wealth is equal to the sum of sellers’ payoffs, consumers’ payoffs, and NGOs’ payoffs, minus the number of sold goods times the minimal production cost that is 20. The total wealth per period, from period 6 to 20, is on average equal to 3400 ECU. Table D1 presents the average sum of sellers’ profits, consumers’ payoffs, donations to NGOs and production costs per period, from period 6 to 20, in each treatment. The number of sold goods on the market is also presented. In table D2, we report variations of payoffs of sellers, consumers, NGOs and of production costs when a label regulation is introduced in comparison to a situation with no label regulation. NS Treatment L Treatment C Treatment CR Treatment Sellers’ profits 502.30 477.98 544.03 483.93

20  

25  

30  

35  

1   2   3   4   5   6   7   8   9   10  11  12  13  14  15  16  17  18  19  20  

Average  price  on  the  market  

Periods  

Unlabeled  good  -­‐  NS  Tr.  

Unlabeled  good  -­‐  L  Tr.  

Labeled  good  -­‐  L  Tr.  

20  

25  

30  

35  

1   2   3   4   5   6   7   8   9   10  11  12  13  14  15  16  17  18  19  20  

Average  price  on  the  market  

Periods  

Unlabeled  good  -­‐  C  Tr.  

Labeled  good  -­‐  C  Tr.  

Unlabeled  good  -­‐  CR  Tr.  

Labeled  good  -­‐  CR  Tr.  

34

Consumers’ payoffs 1893.17 1828.09 1832.62 1 910.96 Donations to NGOs 22.53 76.38 28.23 21.11 Production costs 982.00 1017.56 995.11 984.00 Number of sold goods 49.10 50.88 49.76 49.20

Table D1. Sum of payoffs per period per treatment

NS Treatment L Treatment C Treatment CR Treatment Sellers’ profits - -24,32 +41,73 -18,37 Consumers’ payoffs - -65,08 -60,54 +17,79 Donations to NGOs - +53,84 +5,70 -1,43 Production costs - +35,56 +13,11 +2,00 Number of sold goods - +1,78 +0,66 +0,10 Table D2. Transfers of sum of payoffs per period per treatment relatively to the NS treatment