最后通牒博弈是探究公平问题的经典博弈。1人对1人最后通牒博弈的个体决策机制已有很多研究，研究发现公平社会偏好（如嫉妒、内疚等）是个体决策机制的重要因素，公平社会偏好导致了个体行动偏离纳什均衡。然而，现实社会中个体往往是在群体情景中进行决策的，群体中他人的信息可能会影响个体的决策。因此，本文的研究问题是探讨群体重复最后通牒博弈中的个体决策机制，重点分析群体信息如何影响个体决策。

本文针对该研究问题设计了群体重复最后通牒博弈规则，其中群体结构设计了提议者和响应者之间全连边和动态连边两种交易结构。交易策略集设计为只包含相对公平和相对不公平两种策略的迷你博弈形式，以更准确地体现出个体在决策中考虑的公平偏好。基于该博弈建立了刻画个体决策机制的学习模型，针对群体信息对个体决策的影响提出了三个核心问题：1）个体如何基于群体信息进行学习？2）群体信息如何影响个体的公平偏好？3）个体如何考虑自己的信息作为群体信息产生的影响？

基于该博弈的群体最后通牒博弈实验数据对模型参数进行估计和检验，回答了三个核心问题：1）个体基于群体历史行动信息形成信念，并根据自身信息可能影响他人行动修正信念；2）群体信息诱导个体产生嫉妒和内疚公平偏好效用；3）个体行动信息会作为群体信息对他人行动产生影响，个体据此建立声誉。由此得到结论：个体在群体重复最后通牒博弈中进行策略决策时，群体信息对个体决策的影响主要由三个核心机制构成，即信念学习、公平偏好和建立声誉。

最后，本文基于该模型进行了多主体仿真模拟。包括两部分：1）重现实验中的个体决策过程。检验模型模拟个体决策行动随时间的演化结果与实验中的个体行动数据的一致性，进一步验证模型能够真实刻画现实中的个体决策机制；2）模拟核心决策机制。将模型中公平偏好与建立声誉机制的参数在合理参数空间中遍历，分析其对公平结果的影响，结论为：群体信息诱导提议者产生的嫉妒偏好会导致不公平，产生的内疚偏好会促进公平。群体信息诱导响应者产生的嫉妒偏好会促进公平，产生的内疚偏好在全连边下会导致不公平，在动态连边下对公平的影响不显著。响应者如果认为自身行动信息作为群体信息会影响对手行动，则会有建立声誉的动机，该动机会促进公平。

Ultimatum game is a classic game to explore fairness. There have been many studies on the individual decision-making mechanism of the one-person ultimatum game. It is found that fair social preferences (such as jealousy, guilt, etc.) are important factors in the individual decision-making mechanism, which leads to individual actions deviating from Nash equilibrium. However, in the real society, individuals usually make decisions in the group situation, and the information of others in the group may affect the decision-making of individuals. Therefore, the research problem is to explore the individual decision-making mechanism in the group ultimatum game, focusing on the analysis of how group information affects individual decision-making.

In this paper, the rules of group ultimatum game are designed to solve this problem. Two kinds of trading structures are designed in the group structure: full and dynamic linkage between proposer and responder. The trading strategy set is designed as a mini-game containing only relatively fair and relatively unfair strategies, so as to more accurately reflect the fairness preference considered by individuals in decision-making. Based on this game structure, a learning model is established to describe the individual decision-making mechanism, and three core questions are put forward for the influence of group information on individual decision-making: 1) How do individuals learn based on group information? 2) How does group information affect individual fairness preferences? 3) How do individuals consider the impact of their own information as group information?

Based on the experimental data of the group ultimatum game, the parameters of the model are estimated and tested, and three core questions are answered: 1) Individuals form beliefs based on group historical action information, and modify beliefs based on the possibility of their own information affecting others' actions; 2) Group information induces individuals to produce the fair preference effect of jealousy and guilt; 3) Individual action information will affect the actions of others as group information, based on which individual reputation can be established. It is concluded that when individuals make strategic decisions in the group ultimatum game, the influence of group information on individual decisions mainly consists of three core mechanisms, namely, belief learning, fairness preference and reputation building.

Finally, multi-agent simulation is carried out based on this model. It includes two parts: 1) Reproduce the individual decision-making process in the experiment. To verify the consistency between the evolution results of simulated individual decision-making actions over time and the individual action data in the experiment, further verify that the model can truly depict the individual decision-making mechanism in reality. 2) Simulate the core decision mechanism. The parameters of fairness preference and reputation building mechanism in the model are traversed in the reasonable parameter space to analyze their influence on fairness results. The conclusion is that the jealousy preference of the proposer induced by group information will lead to unfairness, and the guilt preference will promote fairness. The jealousy preference of the responder induced by group information can promote fairness, and the guilt preference can lead to unfairness under full connection, but has no significant effect under dynamic connection. If responders believe that their own action information, as group information, will affect the proposers' action, they will have the motive to build reputation, and this motive will promote fairness.