- Moris Triventi: "Unfair Educational Inequality - Across the World: a Theoretically Driven Machine Learning Approach"
A society with a strong relation between individuals' ascriptive characteristics and educational achievement challenges the equity and efficiency of educational systems, since it tends to misallocate human capital and it penalizes talented individuals born in disadvantageous conditions.
This work aims to provide a theoretically informed analysis distinguishing total inequality from unfair inequalities in students' academic competencies, relying on John Roemer's (1998) normative theory of equal educational opportunity.
We draw information from a pooled large dataset comprising all waves of the OECD Program for International Student Assessment (PISA) across 99 countries worldwide from 2000 to 2018, with an overall simple size exceeding 3,5 million students.
We use student test scores in mathematics as outcome, and a series of variables capturing individual circumstances (gender, mother's and father's education and occupation, geographical area and migration background) as predictors.
We develop a data-driven approach obtaining the partition in student types with transformation trees, a learning algorithm recently introduced by Hothorn and Zeileis (2021), which is trained to predict test scores allowing the set of circumstances to affect expectation and higher moments of the outcome distribution.
By this way, we move beyond an approach solely focused on average achievement to a more comprehensive one taking into account the whole achievement distribution and potential complex interactions among individual circumstances in determining inequality of educational opportunities.