Three essays on social dilemmas with heterogeneous agents
With new tools such as evolutionary game theory and agent based modeling this, dissertation expands the traditional model of human behavior within the BPC model. All three essays involve heterogeneity. All model relatively simple agents playing a common social dilemma repeatedly with other randomly chosen agents from a large population. In the first essay agents come to their beliefs in different ways. In the second their social preferences differ. The third explains how heterogeneity could result given a new version of the public goods game with punishment. All three can predict high levels of cooperative behavior. This dissertation explores the state of the art in three areas.
Chapter 2 models a heterogeneous population of artificial agents playing the Trust Game. Individuals learn which strategies yield the higher payoff according to one of a number of different learning rules. Given homogeneous agents, it is not likely that the most widely used learning models in economics will lead to much if any trustworthiness except by mistake. We show that heterogeneous learning can predict significant levels of trustworthiness and trust when the popular optimizing learning types are joined by Pavlovian and Conformist learners.
In the model presented in Chapter 3 agents are heterogeneous in terms of their social preferences. I first review the literature on self and other regarding preferences revealing the deficiencies in existing models. Then I simulate mini punishment games with heterogeneous preferences. This model allows for selfish, altruistic, reciprocal, spiteful and egalitarian preferences using different parameterizations of the same function. With homogeneous preferences of any type the model predicts zero punishment and either near unanimous cooperation or defection. Alternatively, the coexistence of significant punishment, contributions and defection may be explained with heterogeneous agents.
Chapter 4 presents a version of the public goods game with punishment. Socially costly punishment maintains cooperation and is shown to be stable in a single stage N-player game with mutation. In this game effort exerted during punishment is effort that would otherwise be available for primary public goods provision. This shows a reciprocity norm could evolve in large groups given selfish agents without the benefit of reputation.