Font Size: 

The aim of this work is to analyse and compare PISA2015 results in mathematicsin nine world countries, finding out which are student and school levelscharacteristics related to students’ performances. Based on the fact that education systems are different across countries, the main methodological issue is to use flexible methods that do not force any functional relationships between the variables. We therefore apply tree-based methods in a two-stage procedure: in the first stage, random effect regression trees are used in order to relate student performances to students’ characteristics and to estimate school-value added; while in the second stage, school value-added is related to school level characteristics by means of regression trees and boosting. Results show that three-based methods well fit the problem, being able to explain a good part of variability and identifying different significant features across countries.