Speaker: Ángel Simón Marrero Llinares. Assistant Professor of the Department of Economics, Accounting and Finance - Universidad de la Laguna (Santa Cruz de Tenerife, España).
Title: “Predicting Educational Performance: Bridging Algorithmic Fairness and Inequality of Opportunity”
Abstract: This paper explores the intersection of Inequality of Opportunity (IO) and AI fairness in predicting academic performance. While AI fairness metrics, such as Equality of Opportunity, aim to ensure unbiased predictions, they do not always align with the socioeconomic concept of IO, which defines unfairness as inequalities caused by factors beyond individual control (i.e., circumstances). Using a dataset of primary school students to predict end-of-primary academic performance—including failure risk and excellence—we compare three models reflecting different normative principles: (1) a model incorporating both early- primary academic performance and individual circumstances; (2) a model using only past performance; and (3) a model using only the component of past performance that is uncorrelated with circumstances. Models (1) and (2) generate biased predictions that favor certain groups, as they fail to disentangle individual effort from inherited circumstances. We show that this bias relates to the degree of IO in society and how each circumstance drives it. In contrast, model (3) achieves a better balance between predictive accuracy and AI fairness. However, fairness gains under this model are rarely Pareto improving, as they often involve redistributions of predictive accuracy across groups, and their translation into improvements in IO depends on the policy objective. These findings provide insights for policymakers designing equitable AI models in education, helping ensure that AI-driven educational assessments mitigate, rather than reinforce, inequality of opportunity.
Discussant: Pedro Salas Rojo. Lecturer of the Department of Economics - CUNEF Universidad (Madrid, España).
Tuesday, March 3, from 12:00 to 1:00 p.m. (Google Meet).