Regression Trees in Education
Thu, 09 Feb 2017 13:33:00 GMT
Economics Research Group Seminar
Speaker: Chiara Masci – Politecnico di Milano
Wed, 15th February 2017, 13:15, BSG/28
All staff and students welcome
Regression Trees in Education: Explaining the Value Added of Schools in a Flexible Way
The main objective of the work is to explain how the Hungarian schools' efficiency is driven by school and district level characteristics. The educational system has a hierarchical structure: students, classes and teachers, sites and directors, schools and school principals. Student achievements are influenced by all the levels and, especially, by how these levels interact among them. Our aim is to analyze the educational process, considering its nested structure and without imposing any further functional form. Most of the models used in the literature to explain school value-added are based on parametric assumptions and linear relations between the answer variable and the predictors. Our approach is based on a two-stage analysis: (i) in the first stage, we estimate the sites' efficiency scores by means of Data Envelopment Analysis (DEA), (ii) while in the second stage we use Regression Trees and Random Forests to relate the estimated efficiency scores to site and school level characteristics. In this perspective, tree-based methods are worth, since they do not force any specific functional relationships between the variables and, especially, they allow the variables to interact within and between the levels.
