A0258
Title: Heterogeneous longitudinal structural equation modeling and variable selection
Authors: Chuoxin Ma - Beijing Normal-Hong Kong Baptist University (China) [presenting]
Abstract: Motivated by the China Health and Retirement Longitudinal study, the primary focus is the association between depressive symptom trajectories and various direct and indirect factors. Since the surveys were taken from about 10,000 households in 150 counties and 450 villages, there may be a substantial heterogeneous grouping effect in the longitudinal measurements of the variables. To address these features, a method is proposed for simultaneous mediator selection, subgroup detection and modeling of the longitudinal trajectories. Longitudinal structural equation modelling (LSEM) with regularization was employed to select the important mediating variables and confounding variables and explore the heterogeneity of the direct effects and mediation effects among different groups. The accuracy of the subgroup detection and variable selection were evaluated in numerical studies. The real analysis suggested that the population can be divided into four groups, and cash assets and medical expenses were important mediators between education level and depressive symptom trajectories.
