Multilevel modeling as a tool for student guidance
Leonardo Grilli & Carla Rampichini
 
Abstract
 
High school graduates planning to enroll in University need information on their chances of obtaining a degree in a reasonable time and their occupational opportunities after degree. These chances depend both on student characteristics and university quality.
Since the analysis of the educational process is difficult, the quality of the universities is usually measured via an input/output approach, where the process is a sort of black-box, and the output (outcome) is evaluated in the light of the input.
The university outcomes can be measured only through the consequences on the enrolled students, so a given university may produce different outcomes depending on student characteristics. Therefore, any fair comparison among universities needs an adjustment for the features of the enrolled students, thus student level data. Since the students are nested in schools and schools are nested in universities, the data have a hierarchical structure.
Given the hierarchical structure of the data and the categorical nature of the outcomes considered, the analysis is usually performed in a multilevel framework via Generalized Linear Mixed Models (GLMM).
This presentation illustrates the topic, outlining the state of the art and stressing potentialities and limitations of the statistical methods currently used.