Title: Recursive partitioning of clustered and longitudinal data with GLMM trees
Authors: Marjolein Fokkema - Leiden University (Netherlands) [presenting]
Abstract: Subgroup and moderator detection is of interest in many research fields, e.g., to find subgroups that show differences in treatment effects, or differences in growth trajectories over time. Recursive partitioning or decision-tree methods are pre-eminently suited for this task. Often, researchers may want to detect subgroups or moderators in clustered or longitudinal data. Several existing recursive partitioning methods allow for detecting subgroups in clustered and longitudinal data, like SEM trees, RE-EM trees and GLMM trees. These methods estimate a recursive partition, while accounting for dependence between observations through the estimation of random effects. At the same time, the methods differ in terms of model specification and estimation procedures. Differences and similarities between the methods are discussed. Furthermore, the focus will be on the different ways in which GLMM trees can be specified, and how characteristics of the data problem can best be accounted for, like the measurement level of the partitioning variables, or the strength of the random effects. Simulation results will be presented, and a real data example on subgroup detection in children reading trajectories will be used to illustrate the effects of different model specifications and estimation procedures.