Séminaire GREDEG : Gabriel BAYLE (CEE-M - Montpellier)

This paper introduces a novel bottom-up approach to identifying distributional preferences among individuals using experimental data and unsupervised machine learning techniques. By employing the Dictator Game (DG) and the Ultimatum Game (UG), we observed significant behavioral heterogeneity. Our K-means clustering algorithm revealed four distinct behavioral types: ``Strict Relativist'' (25%), ``Self-interested'' (16.3%), ``Moderate Relativist'' (20.1%), and ``Greedy'' (38.1%). These types were defined using two dimensions: inequality aversion and behavioral symmetry. Our approach addresses several limitations of existing methods by avoiding predefined assumptions and providing a robust and scalable framework. We studied the validity of these types through an additional studies, showing their impact on contribution behaviors in a Public Good game and exploring socio-demographic determinants. Our findings demonstrate the feasibility of this method for capturing preference heterogeneity in diverse populations and its potential applications in policy design and economic modeling.