Essays on cooperation

Huanren Zhang, Purdue University

Abstract

Cooperation between people with different specializations is the driving force behind economic development. While cooperation is mutually beneficial to all participants, it requires one to give up her immediate gain for the benefit of others. This tension between self-interest and the interest of others poses a social dilemma, making it difficult to explain why people cooperate. This dissertation consists of essays that study people's cooperative behavior. For the most part, this dissertation adopts an evolutionary perspective to understand how different factors influence cooperation using approaches including agent based modeling and laboratory experiments, complemented by theoretical analysis. Chapter 1 introduces existing theories on the evolution of cooperation and briefly covers methodologies used in this dissertation. In Chapter 2, a computational experiment is designed to study how the possibility of making errors influences cooperation. A finite number of agents are paired with each other to play the indefinitely repeated prisoner's dilemma, where actions are implemented with noise. Each agent chooses a strategy represented by a finite automaton and play against other agents. Agents are boundedly rational; they update their strategies by imitating other successful agents. This imitation learning makes it more likely for successful strategies to spread in the population. A genetic algorithm is used to model the imperfect learning process, where complex strategies are more likely to be distorted and thus implementation costs are implicitly imposed. The results show that a small amount of noise can increase overall cooperation and the stability of cooperation when the benefit of cooperation is sufficiently high. An investigation of the strategies used by the agents reveals that, consistent with the theoretical analysis, the Tit-For-Tat strategy leads to the emergence of cooperation, but in the presence of noise, the error-correcting Win-Stay Lose-Shift strategy plays an important role in sustaining cooperation. In Chapter 3, I use agent-based modeling to study the emergence of conditional cooperation in the presence of both individual-level and group-level selection. On the individual level, agents play public goods games using conditional strategies represented as piecewise linear response functions; on the group level, groups engage in conflicts with a certain probability. In contrast to previous studies, I consider continuous contribution levels and a rich set of conditional strategies, allowing for a wide range of possible interactions between strategies. The simulation results show that the existence of conditional strategies makes it possible for cooperation to emerge. The strategy that eventually dominates in the population exhibits two properties: (1) It is unexploitable with strong individual-level selection; (2) It can achieve full contribution to outperform other strategies in the group-level selection. The success of this strategy is robust to the initial conditions as well as changes of important parameters. The success of this strategy is robust to initial conditions as well as changes of important parameters. I also investigate the influence of different factors on cooperation levels, including group conflicts, group size, and migration rate. Their effect on cooperation can be attributed to and explained by their influence on the strength of individual-level and group-level selection. While the previous two essays use computational modeling, Chapter 4 is an essay that uses laboratory experiment to study cooperative behavior. In order to study how risks can influence cooperation, I introduce two types of risks into a public goods game: risk that is common among all group members (the COM treatment), and risk that is independent across individuals (the IND treatment). In both treatments, zero contribution is the only equilibrium. In the experiment, contributions to the public goods approach zero under independent risk, while cooperation remains at a high level when common risk is present. The treatment effect is consistent with the quantal response equilibrium on the aggregate level. Analysis of the data provides strong evidence that people are conditional cooperators. A large variation in group contributions is observed within the COM treatment. Further analysis shows that initial beliefs can explain most of the variation in the long-run contribution level, implying the importance of beliefs in establishing group norms.

Degree

Ph.D.

Advisors

Romero, Purdue University.

Subject Area

Economic theory

COinS