| Summary: | Despite the popularity of the notion that social cohesion in the form of dense social networks promotes cooperation in Prisoner's Dilemmas through reputation, very little experimental evidence for this claim exists. We address this issue by testing hypotheses from one of the few rigorous game-theoretic models on this topic, the Raub & Weesie model, in two incentivized lab experiments. In the experiments, 156 subjects played repeated two-person PDs in groups of six. In the "atomized interactions" condition, subjects were only informed about the outcomes of their own interactions, while in the "embedded" condition, subjects were informed about the outcomes of all interactions in their group, allowing for reputation effects. The design of the experiments followed the specification of the RW model as closely as possible. For those aspects of the model that had to be modified to allow practical implementation in an experiment, we present additional analyses that show that these modifications do not affect the predictions. Contrary to expectations, we do not find that cooperation is higher in the embedded condition than in the atomized interaction. Instead, our results are consistent with an interpretation of the RW model that includes random noise, or with learning models of cooperation in networks.
|
|---|
