A New Multilevel Classification and Regression Tree Approach for Binary Data With a Hierarchical Structure

• In Event: 34.079 - Roundtable Session 14
  In Roundtable Session: 34.079-11 - Ordinal and Binary Data: From Differential Item Functioning to Classification and Regression Tree

Abstract

This study presented a new multilevel CART approach, which combined the single level CART with the multilevel logistic regression under the framework of expectation-maximization (EM) algorithm to model hierarchical binary data. Results showed that this multilevel CART approach provided substantial improvements on reducing root-means-square errors and misclassification when comparing with single level CART, multilevel logistic regression, and single level logistic regression. This advantage of using multilevel CART approach was consistently found across various multilevel model types, sample sizes and intra-class correlations.

Author

• Shuqiong Lin, Texas A&M University - College Station