Title: Recursive partitioning methods for prediction in education: application to the identification of students at-risk for academic failure

Authors: W. Holmes Finch

Addresses: Department of Educational Psychology, Ball State University, Muncie, IN 47304, USA

Abstract: Prediction of group membership continues to be an area of open and active research in the social sciences, including education. Many methods exist for this purpose, and a number of studies have investigated their properties in terms of prediction accuracy. However, results of simulation studies as well as applied work continue to demonstrate that extant methods do not provide optimal predictions in a number of situations, particularly those involving non-linear relationships, and when predictor variables interact with one another. The current study seeks to address these issues by describing the use of recursive partitioning methods for group prediction. These techniques, which include classification and regression trees, random forests, and boosted regression trees, show promise in a number of areas but remain relatively little used in the social sciences. In this paper, each method is described in some detail and then applied to the problem of predicting success and failure on a standardised achievement test using a formative assessment and selected demographic variables. Results show that the recursive partitioning methods provide more accurate predictions and more information about which variables contribute to prediction accuracy than do logistic regression or generalised additive models.

Keywords: recursive partitioning; at-risk identification; classification trees; regression trees; random forest; group membership prediction; education; at-risk students; academic failure; formative assessment; demographic variables; prediction accuracy.

DOI: 10.1504/IJQRE.2014.064391

International Journal of Quantitative Research in Education, 2014 Vol.2 No.2, pp.133 - 152

Received: 26 Apr 2013
Accepted: 16 Dec 2013
Published online: 30 Aug 2014 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article