Greenwashing and enforcement

In this work, we investigate the role of different levels of enforcement, and specifically the probability of detection:
- in altering profitable conditions to do greenwashing and the consequent behaviour of sellers;
- in buyers' willingness-to-pay for products with green claims;
- in buyers' actual purchase actions.

The treatments are focused on the role of detection probability in (not) deterring greenwashing from the sellers’ side and in shaping buyers’ revealed preferences for green products.
The detection probability chosen depends on the theoretical model described in the Docs of the present Trial registration. In general, beyond the actual parameterization, they are chosen to accomplish two goals: a) examine individuals’ behaviour in both equilibria spaces and b) enable participants an undemanding understanding of the condition of enforcement.
The resulting levels of ϵ are 1% for the Strong Pooling treatment, 20% for the Weak Pooling treatment and 50% for the Separating treatment.

2024-01-08

2024-02-29

The outcomes are divided into two groups depending on whether they refer to the sellers or the buyers.

The main outcome variables to test sellers’ behaviour are: a) the choice to greenwash while possessing a conventional
product; b) the stated expected demand for both conventional and green products.
The first variable is obtained by looking at the variable greenwashing which is gathered directly through oTree (in Part
A) when a player with a conventional product decides to opt for a green claim. The variable takes the value ”Yes”/”No”
which can be easily re-constructed as a dummy taking value 1 if there is greenwashing and 0 otherwise.
The second main variable is the expected demand which is generated in Part C of the experimental procedure. Sellers
are asked with an economic incentive to reveal what they think is the demand for both green and conventional
versions of each product. The choice is among a set of 10 alternatives with ranges of 10% (0-10%,11%-20%,...,91%-
100%), meaning that participants have to pick the range within which they think the demand for that specific product
lies.
Buyers’ main outcome variables are the willingness to pay for each product displayed and the actual choice to purchase
or not the green product. Both variables are gathered through oTree in Part A. The WTP and choice are filtered
through the variable final claim which reports the claim (potentially) after authority intervention. Their nature is
very straightforward as is displayed in the summary table 4.
Additionally, buyers reveal what they think is the behaviour of sellers holding a conventional when facing the choice
to/not to greenwash. This choice is expressed in Part C and has two outcomes "Yes”/”No” for each product which
can be easily converted into a dummy taking value 1 if they believe that a conventional seller does greenwashing in
such conditions or 0 otherwise.

A variable eprofit is generated for the sellers, to compute the expected profit from the expected demand.
This value is generated for each product seen by each seller and it is equal to:
(expected demand claim B * price B - d)(1-ϵ) + (expected demand claim A * price A - f - d) ϵ

This means that the expected demand for a product with claim B is multiplied by the price relative to the price B minus the disclosure cost d and everything is multiplied by the probability of non-detection (1-ϵ). This amount is summed with the potential loss (or profitability) of being sanctioned: the expected demand of claim A times the price of product A (because the authority changes the claim of the monitored greenwashing) minus the sanction size f and the disclosure cost d. This is multiplied by the probability of receiving the sanction ϵ.

Secondary outcomes are:
- the risk aversion index from the Bomb Risk Elicitation Task (BRET)
- interest and experience in purchasing products
- a series of variables about environmentalism and climate activism
- trust in green advertising
- trust in institutions regulating misleading advertising
- beliefs about greenwashing and its enforcement in real life
- demographics (age, gender, job, field of study, sustenance, income)

The design is the reproduction of the theoretical model in practical terms. There are two types of players, a seller and a buyer. The seller sells a good with or without credence attributes. Product type g is a green good, namely a good with the green credence attribute while type c is a conventional good without that attributes. The other product characteristics are identical. Only the seller possesses the information about the environmental performance of the good. Buyers cannot obtain the information on their own through experience or search.
The seller can choose the claim and consequently, the price to send to the buyer. The set of possible prices is reduced to two levels for simplicity: the price for a good with green claim Pg and the price of a conventional good Pc. If the conventional good has a false claim, its price is Pg which is strictly greater than Pc. On top of this, the seller can signal the quality of the product with a claim about the product’s sustainability. Adopting a signal costs D which is small and which is identical both for truthful and false claims.
By adopting a false signal, the seller incurs in the probability of being caught by an authority which with ϵ probability verifies the signal truthfulness and applies a fine f of a fixed and known amount.

Game procedure
At the beginning of the study, once the instructions have been read, the seller is exogenously given a product which can have a 50% probability of being green and 50% probability of being conventional. The seller knows the actual characteristics of the product displayed on the screen. In the same page, the seller decides whether to adopt a sustainability claim and incur cost D or not. The choice of the signal influences the price at which the good is sold: a good with the green claim is sold at price Pg; a good with no claim at Pc. The seller with the conventional product
has a binary choice between “Claim A” and “Claim B” where “Claim B” is the green (false) option. The seller with the green product instead can only choose ”Claim B” which is, in practice, true. Once the seller has made the decisions the so-called “market authority” randomly checks some of the product types and the claims adopted. The probability differs across treatments (see section 3. The authority verifies automatically
the coincidence between the claim and the product type with ϵ probability and applies a fine f if a green claim has been adopted on a conventional product. Simultaneously, in the case that a false claim is detected, the authority restores the correct price of the good Pc.
In the last stage, the buyer receives the product on his/her screen and elicits two decisions. Firstly, the buyer sees only the claim after possible authority checks and elicits their willingness to pay for the product. Secondly, the buyer is informed about the products’ market prices and decides whether to purchase the product or not by paying the relative price.
Once the buyer has taken both decisions, pairs are randomly shuffled and participants are informed that they might be matched with another partner in the next round.
The actual verification is not public knowledge: sellers discover about the sanctions received only at the end of the study, while buyers never discover that.

Preliminary works
This design has been supported by preliminary work to establish a design as fine-tuned as possible considering the multiple factors affecting individual purchasing decisions. Sections 4.1.2 and 4.1.2 report a summary of the goals and outcomes of the online survey held in March 3rd 2023 and the laboratory auctions on products held between October 24th and November 3rd 2023.

Online survey
The first supporting exercise was an online survey (n=178) to test a set of products both in green and conventional versions to elicit hypothetically the WTP for both conventional and green versions of the products. The set of products chosen was composed of dark chocolate, coffee, avocado, cola, dish-washing liquid, cotton buds, toothpaste, soap and basmati rice. All these products have a set of different characteristics that generate heterogeneity and, at the same time, the nature of credence attributes. Heterogeneity allowed me to verify the existence of a WTP gap between green and conventional products. Additionally, the online experiment allowed me to pilot testing the role of
image, the potential presence of order effect and, finally, the role of the verification of the products on WTP. Most of the results are statistically insignificant since the 2x2x2 design reduced the treatment sample size. However, this online survey allowed me to understand the noise and salience of images and the presence of order effect. As a consequence, in the following study products’ images were removed and I decided to show one of the two versions of the products to the participants (green or conventional). Finally, the online survey allowed me to establish the maximum WTP students would pay for the products to be used in the later Becker-Degroot-Marshak during laboratory practices.

Laboratory auctions with BDM
After the online survey, a series of laboratory auctions (n=85) have been deployed to elicit in real terms the WTP for green and conventional products in similar conditions as in the following Game.
The methodology adopted is the Becker-Degroot-Marshak following the approach supported by Bohm et al. (1997): participants (as buyers) must elicit a bid for each product without knowing its price which is contained in an envelope. The price has been drawn from a uniform distribution taking values from 0.05 € to the maximum WTP elicited during a “previous study” (i.e. the online survey). Bidders are informed of how the upper bound is generated but they do not know the amount.
This auction helped to elicit the WTP to establish Greenwashing Games’ sellers’ prices fixed at the median level of the auction. This choice enables the examination of partial equilibria because it sets the a priori distribution of expected demand equal to 50%. Additionally, during these auctions, participants were informed that 100% of sustainability claims had been verified and they were correct, providing the upper bound for WTP for such products conditional on verification.
Each participant was endowed with 10€ for each round to express his/her WTP for the randomly displayed product. Only one round’s decision was implemented for the payment and the distribution of the product at the end of the study. The payoff is 10eminus the envelope price and the product if the WTP is greater or equal to the product envelope price or 10 € otherwise.
The reported Table 1 (see attachments) displays the mean and median of the WTP both for conventional and green products. Overall we observe a green premium for most products in both statistics with the single exception of basmati rice which displays only a minor premium among the observed population. As a consequence, basmati rice has been removed from the list of products of the Greenwashing Game given the lack of monetary incentives to greenwashing for sellers.

Greenwashing game parameters
The parameters chosen for the Greenwashing Game are essential to bring the theoretical predictions to practical terms. In particular, the ϵ∗ is computed starting from the knowledge of prices Pg and Pc; F(γ) as the probability of selling g; H(σ) as the probability of selling c; D: the green advertising cost; f: the sanction size.
Prices and probabilities of selling both products are established by design. The prices are set at the median level obtained through BDM auctions so that the probability of selling g and c is also equal to 50%.
For what concerns the green advertising cost, and the sanction size we refer to Table 2 (see attachments) to show the estimated detection probability ϵ∗, the key element of the study. By fixing the green advertising cost to 0.05€ and varying the sanction size I was able to detect a sanction level that creates useful conditions for testing. The sanction size decided is 60 cents because in this range we obtain heterogeneity of ϵ∗ within two boundaries 16% and 54%.

Several randomization procedures are adopted. Here is the description:

Participants are randomly allocated to the seats (or to turned-away reserves) through a visible random draw when they enter the laboratory.
The software oTree randomly assigns participants to a participant number independently of the seat number they are working.
The software randomly assigns the roles of sellers and buyers.
The first product is fixed for econometrics considerations, while from round 2 participants receive a product assigned to that participant number. In every session, the product-participant assignment varies randomly.
The product assigned to sellers can be green or conventional, and this is the result of a random draw.
The prices contained in the envelope are randomly drawn before each session starts through Microsoft Excel to facilitate the envelope procedures.
Each buyer can receive a different product, which is randomly drawn before the session starts. The software Excel draws a product for each participant number. This is chosen to reduce the number of products brought to the laboratory for every session.
The sellers' two valid rounds are drawn within the software oTree during the session, and they are different for each participant.

Individual.

No

Individual clusters: 120-160 participants per treatment meaning 60-80 per type.

Minimum: 2 treatments only, 120 participants each, 7 observations per participant: Planned number of observations: 1680 Maximum: 3 treatments, 160 participants each, 7 observations per participant: Planned number of observations: 3360

120-160 participants per treatment arm. Plus 85 participants from the BDM auction already deployed.

By adopting the a priori power analysis to compute the required sample size, we employed a statistical test from the family of t-test for means to analyse the difference between two independent means. Since BDM and pilots had two different sample sizes, we employed the method with different sample sizes to compute the effect size. The mean of WTP for green products for the BDM is 2.65 (SD 1.70) and for the pilot is 2.42 (SD 1.08). The effect size (opting for the lower bound of SD) is 0.21, which is employed together with α = 0.05 and β = 0.80 to compute the total sample size. The result is 558, meaning that each treatment must have 279 observations. This is mostly aligned with the data gathered for the BDM, which counts 285 observations of product WTP for green products. The results are similar when employing a Mann-Whitney statistical test, for which the effect size rises to 0.25 and the resulting total sample size required is 424 observations. These results mean that around 280 observations of product WTP are needed to have a statistically valid result, which translated in number of participants means 40 participants seeing a green product. As a consequence, since some participants see conventional products, the number of buyers should be around 80 people per treatment.