Commitment Game

2022-07-06

2022-12-31

Whether the various promise mechanisms can induce buyers to keep promise and leave feedback, and how does it affect the buying and shipping decisions as well as feedback.

The definition of terms in primary outcomes (promises, shipping, etc.) can be found in experimental design.

The proportion of efficient trades during 10 periods under different treatments.

The definition of terms in secondary outcomes (buying, shipping, etc.) can be found in experimental design.
Efficient trade means that buyer buys the product and the seller ships the product in one time of the trade.

[Design]
There are two treatments and a baseline in the experiment.
The baseline is a buyer-seller trust game modified from that in Bolton et al. (2004). The buyer first chooses whether to buy a unit of product from the matched seller, and the seller decides whether to ship the product (choosing "not ship" will bring additional benefit to the seller and seriously hurt the buyer). After knowing the seller's action, the buyer needs to decide whether to leave costly feedback to the seller that can be seen by all subsequent buyers matched with the seller.

- Treatments 1 and 2 each add a mechanism regarding buyers' commitment (non-binding) to providing feedback, but with different eliciting processes.

- In Treatment 1, the buyer's commitment is elicited by the system (third party), in which the buyer is asked to choose Yes or No on sending a pre-determined message to the seller that committing to leaving feedback along with their buying decision. Note that the commitment is non-binding, which means the buyer's final feedback decision is independent of their commitment. Also, the seller will be informed when the buyer chooses not to send the message.

- In Treatment 2, before the buyer's buying decision, the seller needs to decide whether to ask the matched buyer for committing by leaving feedback after the trade, which means the buyer's commitment is elicited by the counterpart (the seller). Buyers can freely decide whether to reply to this invitation (If the seller asks), and the seller will know the buyer's commitment decisions before they make their shipping decisions. As in Treatment 1, the buyer's final feedback decision is independent of their commitment.

- Each subject is allowed to participate in at most one session of the experiment (i.e.,between-subject design). The data will be collected through a computerized experiment at a University economic laboratory.


[Motivation]
The motivations of the experimental design are fivefold.

- Firstly, it is well studied in the literature that non-binding commitment (or cheap talk) can foster cooperation in various games (e.g. trust game, public goods game， gift-exchange game, etc.), usually through that the trustee makes the direct commitment of cooperation to the trustor. However, when the game is embedded with a reputation system, we consider non-binding commitment may foster cooperation and efficiency in different ways, and it is under explored in the literature. This study can potentially fill the gap.

- Secondly, the reputation system in the online market usually suffers from inadequate provision of feedback and the negative-bias problem. Negative bias problem means that buyer is more willing to leave feedback when there is a negative outcome. Commitment to leaving feedback may help to increase the feedback rate. Moreover, based on the expectation-based explanation of people's keeping promise behavior, a commitment may induce more positive feedback， which may potentially alleviate the negative bias problem.

- Thirdly, it is in the interest of both behavioral and theoretical researchers why people tend to keep their non-binding promises. The two leading explanations are named expectation-based(guilty aversion) explanation and commitment-based explanation. The design in our experiment can provide multiple variations on the buyer's guilty level(including no guilty) on the buyer's keeping promise behavior, which could provide testable hypotheses on these explanations.

- Fourthly, we find the existing literature rarely compare different promise-generating process within the elicited promise (There is literature comparing voluntary promises and elicited promise, for example, see Ismayilov and Potters (2017)). Our study may provide some evidences on this topic.

-Finally, as in some previous literature, the seller can signal their type (or goodwill) by choosing pre-game actions (for example, providing a rebate for feedback (Li and Xiao, 2014), or providing advice on insurance (Grodeck et al., 2022)), and these actions can, in turn, encourage sellers to behave cooperatively. The seller's action of asking for promises in our Treatment 2 can also be viewed as pre-game signaling, and we are interested to examine whether such a signal can be as effective as in other literature.


[Hypotheses]
H1: The feedback rate will be higher when the buyer makes promises in Treatment 1 and Treatment 2 than in the baseline.

H2: The negative bias problem will be alleviated in Treatment 1 and Treatment 2 compared to in baseline.

H3: Given committing, the buyer's guilt of breaking promises is different in Treatment 1 and 2. As a result, the feedback rate will be different in Treatment 1 and Treatment 2. We expect both directions of this comparison is possible, depending on how buyers interpret sellers' asking for promise in Treatment 2. Specifically, if the buyer considers the seller's asking for a promise indicates their direct expectation of the buyer's reciprocal behavior (if the seller ships the product), then the buyer should be guiltier in Treatment 2 when breaking the promise. On the other side, if the buyer considers the seller's asking for promises means nothing other than persuading the buyer to purchase the product, then the buyer should feel less guilty in Treatment 2 when breaking the promise (because they have bought the product and meet the seller's expectation). Therefore, we do not have a particular hypothesis but empirically explore the difference between Treatment1 and Treatment2.

H4: Because of the higher feedback rate (H1), the sellers will have a higher shipping rate in both treatments 1 and 2 than in the baseline. Moreover, the seller's actions of asking for a promise is a signal of their shipping decision: Compared to Treatment 1, the shipping rate will be higher in Treatment 2 for those sellers who ask for the buyer's promise, and will be lower for those sellers who do not.

H5: Buyers view the seller's asking for a promise as a signal of their shipping decision: buying rate will be lower in treatment 2 than in treatment1 and the baseline if the seller does not ask for the promise. Given the sellers ask for promise in Treatment 2, because of the higher shipping rate (H4), buying rate will be higher in both treatment 1 and 2 than in the baseline, and we expect the difference in buying rate between treatment 1 and 2 is possible in both directions because the difference in feedback rate between the two treatments is possible in both directions (H2).

Reference
Bolton, G. E., Katok, E., & Ockenfels, A. (2004). How effective are electronic reputation mechanisms? An experimental investigation. Management Science, 50(11), 1587-1602.
Li, L., & Xiao, E. (2014). Money talks: Rebate mechanisms in reputation system design. Management Science, 60(8), 2054-2072.
Ismayilov, H., & Potters, J. (2017). Elicited vs. voluntary promises. Journal of Economic Psychology, 62, 295-312.
Grodeck, B., Tausch, F., Xiao, E., & Wang, C. (2021). To Insure or Not to Insure? Promoting Trust and Cooperation with Insurance Advice in Markets. Working paper

The randomization is done by a computer.

There are two levels of randomization in the experiment. The first level of randomization is at the individual level, where each buyer is randomly matched with a seller in every single period. The second level is the group level. Each session consists of 24 subjects, where every 12 subjects (6 buyers and 6sellers) are randomly assigned into an independent group at the beginning of the experiment, and interact with other subjects only within the group for the rest of the experiment.

Yes

We plan to have 18 experimental sessions (6 sessions for each treatment), and each session consists of 24 subjects.

432 subjects (216 pairs of sellers and buyers).

144 subjects (6 sessions) for the baseline
144 subjects (6 sessions) for treatment 1
144 subjects (6 sessions) for treatment 2

Before conducting the experiment, we did a pilot with three sessions corresponding to each of the three treatments (each session consists of two independent 12 subjects groups). Regarding the primary outcomes and the secondary outcomes, we run the power test (we set power = 0.8 and significance level =0.05) for each pair of three treatments. [Primary outcomes:] [The probability of reporting] (1). Treatment 1 vs. Baseline treatment: 12 subjects (6 buyers and sellers in total) are required (2). Treatment 2 vs. Baseline treatment: 12 subjects (6 buyers and sellers in total) are required. [Secondary outcomes:] [1. The probability of shipping ] (1). Treatment 1 vs. Baseline treatment: 328 subjects (164 buyers and sellers in total) are required (to detect the difference between treatments) (2). Treatment 2 vs. Baseline treatment: 210 subjects (105 buyers and sellers in total) are required. [2. The probability of buying] (1). Treatment 1 vs. Baseline treatment: 724 subjects (362 buyers and sellers in total) are required (2). Treatment 2 vs. Baseline treatment: 72 subjects (36 buyers and sellers in total) are required. [3. The proportion of efficient trades] (1). Treatment 1 vs. Baseline treatment: 2204 subjects (1102 buyers and sellers in total) are required (2). Treatment 2 vs. Baseline treatment: 96 subjects (48 buyers and sellers in total) are required. [Planned subjects] According to the results of power tests, and considering our budget constraint, we plan to recruit 144 subjects for each treatment for two reasons: Firstly, we want to ensure the comparisons of some primary and secondary outcomes are salient (e.g. Comparison of the probability of buying: Baseline vs Treatment 2; Comparison of the probability of reporting: Baseline vs Treatment 1, Baseline vs Treatment 2; Comparison of the proportion of efficient trade: Baseline vs Treatment 2). Secondly, given the differences in the remaining comparisons are quite close (require at least 210 to obtain significance), means the treatment differences (Baseline vs. Treatment 1) are not large. We, therefore, decide to collect a similar sample size as treatment 2.