Caroline Liqui Lung: Learning from Endogenous Social Data

Event details

Author:  Caroline Liqui Lung (City St George’s, University of London)

Abstract: People often learn about opportunity from the outcomes of others. Yet the evidence they observe is endogenous: earlier beliefs shape participation and selection, and thereby the outcomes from which later cohorts learn. This paper asks when Bayesian learners can recover current opportunities, and when statistics used to infer disparities can instead generate and sustain the very patterns they seem to reveal. I develop a general framework of Bayesian learning from endogenous social data and characterize its long-run equilibria. Group-specific returns determine how ability translates into success, while information structures determine what later cohorts observe about the process generating those outcomes. I classify information structures by whether Bayesian learning corrects inherited disparities or makes them self-reinforcing. When statistics confound current opportunity with belief-driven behaviour, history can become embedded in the evidence itself, sustaining disparities across social groups long after the forces that created them disappear. This classification identifies information design as a potentially low-cost tool for promoting diversity and reducing persistent inequality, and provides a portable framework for settings in which participation and selection shape the data from which others learn.

Host: James Choy (York)

Cluster: Micro Theory