Dissertation abstract: Contribution to a public good - Theoretical analysis and experimental evidence

The work undertook is located between Public Economic Theory and Experimental Economics. The object of the thesis consists in analysing the aggregate behavior and the individual heterogeneity in a voluntary contribution game. The thesis defended here is that overcontribution in comparison to the Nash equilibrium of the game, can not be explained neither by judgement errors, nor by the information on the individual behaviour of the members in a one given group. The level of contribution observed remains nevertheless inferior to the one simulated using the E.W.A. learning model. The dissertation is composed of three parts and six chapters. The first chapter expresses the various theoretical mechanisms of production of a public good, while the second one presents an overview of the experimental literature using voluntary contribution mechanisms. The second part carries on the introduction of an interior solution in a public good game in order to distinguish an explanation of overcontribution in terms of mistakes or strategies. While chapter three presents the most important works in literature that use an interior solution, the fourth chapter constitutes a personal contribution consisting in an experiment with an interior optimum. Our main result is that individuals contribute a constant part of their social optimum and that overcontribution is not explained by error. We test then the simple learning model R.L. using the observed data on the aggregate level. This model predicts well the observed behaviour. The third part is composed of two experiments where the environment of players is modified. We introduce in chapter five promises as cheap talk and find that they increase contributions at the aggregate level. In chapter six, various conditions of information on individual contributions are tested. The parameter tested is the level of information on “neighbours” contributions given to players. One of the treatments presents full information about individual contributions of the members of the group, while this information is incomplete in the other treatments. Our results show that information has no effect on the level of contribution. We simulate then the EWA learning model both at the aggregate and the individual levels and compare the simulated data to the experimental one. These simulations predict a level of contribution that is higher that the one observed in the experiment. JEL Classification C91, C92, H41 Dissertation Committee: Alan Kirman (Ph.D. Advisor), Université d’Aix-Marseille III Marie-Claire Villeval (Chair), GATE, Lyon, France Jordi Brandts, Universitat Autonoma Barcelona, Spain Charles Noussair, Emory University, Alanta, GA, U.S.A. Sylvie Thoron, GREQAM & Université du Var, France Marc Willinger, Université de Montpellier I, France