Investigating Covariate Effects of Model-Based Recursive Partitioning to Uncover Population Heterogeneity in Longitudinal Study

• In Event: Educational Research From Different Lenses

Abstract

Model-based recursive partitioning is an integration of parametric models into decision tree of machine learning. It has gained growing interests as a complementary data analytic tool to address population heterogeneity in educational research. While the role of informative covariates is extensively emphasized, it has been less examined to investigate the performance of it where there exist interactions effects. A simulation is conduced. Four plausible scenarios are considered for population models based on the specification of both fixed and random effects. Simulation conditions include model misspecification, sample size, effect size of mean differences, types of interaction effects, treatment of covariate types, and pruning strategies. The preliminary results show that the effect size is a key element to recover true subgroups.

Author

• Youngjun Lee, Michigan State University