How Classification Trees Can Be Helpful to Educational Research: A Case Study

• In Event: Poster Session 10
  In Poster Session: Division D Section 2 Poster Session 2

Abstract

Easy to interpret, flexible in application, and inclusive of a variety of variables, decision tree approaches can be very helpful for exploring educational datasets. Tree methods have the potential to reveal interesting higher-order interactions between predictors, often encountered in Education but not easily included in linear model approaches. Yet decision trees are rarely used in educational research. To better understand whether and how tree methods can be useful to researchers, we analyzed a large-scale educational dataset using three tree methods (CART, QUEST and CHAID) versus stepwise logistic regression. While the methods had comparable classification accuracy, tree results provided a richer “overall sense” of the data, and generated unique suggestions about how students’ learning satisfaction can be improved.

Authors

• Emi Iwatani, University of Pittsburgh

• Clement A. Stone, University of Pittsburgh

• Craig N Shealy, James Madison University