Mechanism design uses game theory to analyze how to construct mechanisms that give agents incentives to produce an outcome as optimal as possible for the mechanism designer. My dissertation is on robust mechanism design, where the mechanism designer has a great deal of uncertainty about what agents believe about other agents' preferences and beliefs. Much research on robust mechanism design has focused on finding "dominant strategy mechanisms," where agents' equilibrium actions depend only on their preferences and not their beliefs about the other agents. Focusing on such mechanisms makes uncertainty about agents' beliefs irrelevant. Chapters 2 through 5 of my dissertation explore how effective dominant strategy mechanisms are at achieving mechanism designers' goals. Mechanisms are compared by considering performance of the mechanisms for a wide range of possible agent preferences and beliefs, and then defining one mechanism as an improvement on another mechanism if it performs as well (and sometimes better) for every possible realization of agents? preferences and beliefs. The first two chapters explore this question in the context of a public good problem. Chapter 2 demonstrates that there are mechanisms that improve on the dominant strategy mechanisms in the sense just described; Chapter 3 examines the usefulness of this efficiency concept by providing an example of a mechanism that is both very simple and unimprovable. The fourth and fifth chapters are coauthored with Tilman Borgers and look at the same question for the voting problem, where dominant strategy mechanisms are (random) dictatorships. Chapter 4 shows that random dictatorship can be improved on if the mechanism designer wants to maximize agents? expectation of their welfare, but not ex post welfare. Chapter 5 shows the same results when the designer wants to maximize the sum of two agents? welfare.The last chapter, Chapter 6, is coauthored with Mary Rigdon and reports on an experiment designed to investigate whether seemingly altruistic behavior is actually a response to strategic incentives, by using pre-commitment of some agents to test whether agents act differently when their actions can?t influence others? future actions. Our data support the hypothesis that agents are motivated by social concerns.
