A0666
Title: A novel cluster-weighted multilevel model for two-levels clustering
Authors: Chiara Masci - Universita degli Studi di Milano (Italy) [presenting]
Andrea Cappozzo - Universita Cattolica del Sacro Cuore (Italy)
Abstract: A novel two-level cluster-weighted model designed for hierarchical data is introduced. This approach bridges cluster-weighted modeling with multilevel modeling using discrete random effects, resulting in a powerful framework for two-level clustering. The model identifies latent subpopulations of observations - referred to as profiles - that exhibit heterogeneous characteristics. Within each profile, multilevel models are estimated to capture profile-specific regressions and random effects. At the higher level of the hierarchy, random effects are modeled using a semi-parametric generalized linear mixed model (SPGLMM). The discrete nature of these random effects facilitates the identification of clusters at the top level of the data structure. To estimate the model parameters, an expectation-maximization algorithm is developed, tailored to the proposed framework. The method is validated through a simulation study and an empirical case study. It is particularly well-suited to applications involving nested data structures, such as students within schools, citizens within states, or patients within hospitals. To demonstrate its practical utility, the model is applied to the OECD-PISA dataset to analyze European educational systems. Specifically, the probability of a student being a low performer in mathematics is examined, identifying distinct student profiles and grouping countries according to their impact on student outcomes.
