The Effects of One-Way Feedback and Dialogue on Decision-Making in a Common Pool Resource Game

2025-07-01

2025-07-31

The primary outcome variable is the level of resource extraction by individual participants in each period. Specifically, we will focus on:
1. Average individual extraction per period over the game;
2. Deviations from the Pareto-efficient extraction path;
3. Sustainability indicators, such as whether the group maintains a positive stock of resources over time.

This study will examine how unidirectional and bidirectional dialogues about future resource dynamics will influence individuals’ extraction behavior in a dynamic common pool resource (CPR) game. The experiment will be conducted with Japanese university students, who will be assigned to groups of three and will make repeated decisions on resource extraction over multiple periods.

This study will examine how unidirectional and bidirectional dialogues about future resource dynamics will influence individuals’ extraction behavior in a dynamic common pool resource (CPR) game. The experiment will be conducted with Japanese university students, who will be assigned to groups of three and will make repeated decisions on resource extraction over multiple periods. Participants will be randomly assigned to one of the following communication conditions: (i) unidirectional dialogue initiated by the participant, (ii) unidirectional dialogue generated by a large language model (LLM), or (iii) bidirectional dialogue between the participant and the LLM—either initiated by the participant or by the LLM. The resource dynamics will follow a model incorporating regeneration, and the overall resource path in the game will be pre-designed to be Pareto efficient. We will conduct a post-experiment survey, which included a 10-item Big Five personality test and a 3-item Cognitive Reflection Test (CRT).

randomization done in office by a computer

group level in each session

Yes

300 students

100 groups of 3 participants

300 students (100 groups)

We plan to recruit 300 participants, forming 75 groups of 3. Treatments will be assigned at the group level across four conditions (approximately 18–19 groups per condition). Assuming a standard deviation of 2.0 units and an intra-cluster correlation (ICC) of 0.05, the minimum detectable effect size is approximately Cohen’s d = 0.38 (80% power, 5% significance level), equivalent to a 10–12% difference in mean extraction levels.