The reasons why the Regression Tree Method is more suitable than General Linear Model to analyze complex educational datasets

Abstract

Any quantitative method is shaped by certain rules or assumptions which constitute its own rationale. It is not by chance that these assumptions determine the conditions and constraints which permit the evidence to be constructed. In this article, we argue why the Regression Tree Method’s rationale is more suitable than General Linear Model to analyze complex educational datasets. Furthermore, we apply the CART algorithm of Regression Tree Method and the Multiple Linear Regression in a model with 53 predictors, taking as outcome the students’ scores in reading of the 2011’s edition of the National Exam of Upper Secondary Education (ENEM; N = 3,670,089), which is a complex educational dataset. This empirical comparison illustrates how the Regression Tree Method is better suitable than General Linear Model for furnishing evidence about non-linear relationships, as well as, to deal with nominal variables with many categories and ordinal variables. We conclude that the Regression Tree Method constructs better evidence about the relationships between the predictors and the outcome in complex datasets.