Garbage Can in the Lab

Abstract

We develop an experimental setting where the assumptions and predictions of the garbage can model can be tested. A careful reconstruction of the original simulation model enables us to select parameters that leave room for potential variations in individual behavior. Our experimental design replicates these parameters and thereby facilitates comparison of human behavior with the behavioral assumption of the original model. We find that the majority strategy of human subjects is consistent with the original model, but exhibits some extent of behavioral diversity. What is notable is that, although slightly outperforming the simulation model, human subjects failed to learn at the group level to improve performance. This is because human subjects exhibit fluid diverse behaviors that improve coordination in the face of uncertainty, but hinder collective learning.