An experimental test of the efficacy of a simple reputation mechanism to solve social dilemmas

The evolution of cooperation has been the focus of intense research in the social sciences, natural sciences (especially biology), and even computer science. It has long been recognized that the possibility of future consequences is crucial to the emergence of rational cooperation. It was thought that random matching was isomorphic to one-shot play, but Kandori (1992) showed that a reputation labeling mechanism can be used to support cooperation in the random matching Prisoner Dilemma. We designed an experiment to test this result. We found that while the level of cooperation steadily declined without a reputation mechanism, with our color-coded reputation mechanism the level of cooperation steadily increased with experience. An econometric mixture model consisting of four reputation-conditioned strategies as well as a Level-0 type was fitted to the data. We cannot reject the hypothesis that the majority of subjects used one of these strategies and learned with experience.