Group Learning and Governance

Collaborators: Joshua Cox, Mirta Galesic, Henrik Olsson

Tools & Languages: Python, R, stan, oTree (Django)

Methods: Experiment Design, Reinforcement Learning Models, Hierarchical Bayesian Modeling, Model Comparison, Model Simulation and Model Fitting

🔭 About

When groups make decisions under uncertainty, they must learn not only about their environment but also how to govern themselves. A central trade-off is between fast but sometimes flawed leadership and slower, information-aggregating majority voting. Yet it remains unclear how people learn which governance structure works best, and how this choice depends on uncertainty.

Here we present a real-time, incentivized group experiment paired with a hierarchical reinforcement-learning model to study this process under controlled conditions. Groups of four repeatedly choose among uncertain options while also deciding whether outcomes are set by majority vote or by a leader whose quality varies. We manipulate both task difficulty and leadership quality, allowing us to cleanly separate learning, social influence, coordination costs, and governance choice. The model captures how people learn about option values as well as how they learn which governance rule to use.

Using preliminary data, we examine whether people show a baseline preference for one governance structure, how leadership quality shapes learning and exploration, and whether groups can still coordinate when faced with poor leadership. This framework provides a general experimental and computational approach for studying how leadership and voting shape group learning and coordination under uncertainty.