Charting the Type Space—The Case of Linear Public Good Experiments

Behaviour in economic games is not only noisy. One has reason to believe that heterogeneity is patterned. A prominent application is the linear public good. It is widely accepted that choices result from participants holding discernible types. Proposed types, like freeriders or conditional cooperators, are intuitive. But the composition of the type space is neither theoretically nor empirically settled. In this paper, we leverage machine learning methods to chart the type space. We use simulation to understand what can be achieved with machine learning. We rely on these insights to find clusters in a large (N = 12,414) set of experimental data points from public good games. We discuss ways in which these clusters could be rationalized