Revealed-preference study of the impact of narratives and conceptual frames: evidence from a negative externality laboratory experiment

Experimental conditions in the study vary the conceptual frame of the negative externality. We examine whether subjects’ willingness to sell and buy this “unfair” product is impacted by the conceptual frame.

STUDY 1:
Experimental conditions in the study vary the conceptual frame of the negative externality. In the control (control), the outcomes of the bystander are not reframed for the participants – payoffs associated with choices are simply provided. In the first treatment (reduced), the outcome of the bystander is framed as a reduction in the bystander’s payoff. In the second treatment (profit), the buyer and seller are described as profiting at the expense of the bystander. We examine whether subjects’ willingness to sell and buy this “unfair” product is impacted by the conceptual frame.

STUDY 2:
This study uses a similar set of interventions. In the placebo, participants are given a brief summary of what the Buyer, Seller, and Bystander do in the study. In the byproduct statement, the effects of the market outcome are framed as byproducts. In the profit statement the buyer and seller are described as profiting at the expense of the bystander.

2022-04-09

2022-08-31

STUDY 1:
sell: dummy variable = 1 if seller decides to offer a product for sale
buy: dummy variable = 1 if buyer decides to purchase a product
price: price at which product was purchased

STUDY 2:
offer: whether the Seller offers a product for sale
offer price: the price at which the Seller offers the product for sale
buy: whether the Buyer chooses to buy a product
wtp: the Buyer's willingness to pay
belief_maxwtp: the Seller's beliefs about the Buyer's wtp

STUDY 1
Participants will engage in a laboratory market game with real stakes for themselves and a bystander. Sessions are randomly assigned to one of three experimental conditions - a control and two treatments. In the two treatments, the framing of the negative externality produced by the players is altered.

STUDY 2
Study 2 implements a similar design but using an online asynchronous market on Qualtrics.

See analysis plan.

STUDY 1:
Assignment of conditions to sessions is done non-randomly to ensure even spread across days/times; participants sign up for days/times according to their schedule preferences; payment round is determined randomly by the computer

STUDY 2:
Randomization of condition assignment is performed using Qualtrics' built-in functions.

STUDY 1
experimental sessions

STUDY 2
individual

Yes

STUDY 1
approximately 18 - will depend upon group size and interest at recruitment

STUDY 2
Plan to recruit 2,158 participants

STUDY 1 As many as can be recruited - up to 600 students. STUDY 2 Plan to recruit 2,158 participants - each will make 2 decisions = 4,316 observations

STUDY 1
Approximately 200 depending upon scheduling. If sessions get out of balance due to cancellations, etc., we will err on the side of oversampling experimental sessions.

STUDY 2
T1: 120 Sellers, 96 Buyers, 96 Bystanders
T2: 55 Sellers, 44 Buyers, 44 Bystanders
T3: 55 Sellers, 44 Buyers, 44 Bystanders
T4: 120 Sellers, 96 Buyers, 96 Bystanders
T5: 120 Sellers, 96 Buyers, 96 Bystanders
T6: 120 Sellers, 96 Buyers, 96 Bystanders
T7: 120 Sellers, 96 Buyers, 96 Bystanders
T8: 120 Sellers, 96 Buyers, 96 Bystanders

STUDY 1 Assuming an alpha of 0.01, power of 0.80, a minimum effect size of 0.15 and approximately four predictors, we expect to need approximately 120 participants in our buyer analysis (online a priori sample size calculator for multiple regression: https://www.danielsoper.com/statcalc/calculator.aspx). This would require a sample size of approximately 500 participants due to the study design. We plan to attempt to recruit up to approximately 600 participants overall. STUDY 2 Although we found statistically significant treatment effects (profit vs. control) in Study 1, the study was underpowered due to logistical issues. For example, the effect size when assessing the average offer price (averaged across all regular rounds) across treatment group, we find that the observed effect size was large – d = 2.22. Analysis using gpower demonstrates that our sample sizes in Study 1 allow us to identify a medium effect size (d = 0.5; a difference of approximately 2 experimental points) with a power of just 0.36 at 𝛼=0.01. Assuming similar means and standard deviations in the online study, 96 participants per role per treatment are needed to identify a medium effect size at 𝛼=0.01 and a power of 0.80 (T1, T4, T5, T6, T7, T8). For within-subjects analysis (T2, T3), we need only 43 participants per role to identify a medium effect size at 𝛼=0.01 and a power of 0.80.