| In this thesis, we take a game-theoretic approach to modeling behavior in peer production systems. Our goal in modeling such systems is both to capture the current behavior of the system and to understand how changes in the incentive structure can lead to more desirable system-wide outcomes. We focus on both modeling specific systems as well as general models that are applicable to a wide variety of systems.;The first system that we study is the ESP game. The ESP game is a popular game on the web for labeling images. During play of the ESP game, it seems that users coordinate on easy words. We provide a game-theoretic model that supports this observation. We also study the effect of alternate agent preferences on the equilibrium outcomes and provide a class of reward functions that allow for coordination on more descriptive words as the equilibrium outcome. Our equilibrium analysis focuses on all valuations satisfying a given preference model.;The second system that we focus on is Yahoo! Answers. In Yahoo! Answers, it appears that under the "Best Answer" method of assigning points, users prefer to answer later rather than sooner. We provide a simple game-theoretic model of Yahoo! Answers and present equilibrium analysis that supports this observation. We propose two new methods of assigning points, the Approval Voting Scoring Rule and the Proportional Share Scoring Rule, and show that each can improve the equilibrium outcome. We provide an axiomatic characterization showing that no "reasonable" scoring rule can isolate the best possible equilibrium outcome.;Finally, we study the combinatorial agency problem introduced by Babaioff et al. (2006). The combinatorial agency model is a variant of the classical principal-agent model of economics and is relevant to many crowdsourcing systems. We provide a characterization of the optimal contract as a function of the principal's value for the class of threshold functions. We study the Price of Unaccountability, or the relative loss to the principal due to the strategic behavior of the agents. We compare the Price of Unaccountability across various technology functions and find drastically different results depending on the technology function. |
