A0676
Title: Learning pathways of life events: A sequential allocation Bayesian model with label tracking
Authors: Beatrice Franzolini - Bocconi University (Italy) [presenting]
Andrea Cremaschi - IE University (Spain)
Raffaella Piccarreta - Bocconi University (Italy)
Abstract: The analysis of socio-economic data that tracks life events, such as marriage, childbirth, and employment, can deepen the understanding of representative life trajectories in contemporary societies. However, this typically requires clustering methods for multivariate longitudinal categorical data, and existing model-based methods in the statistical literature remain rather limited in this regard. A flexible Bayesian nonparametric framework is introduced for modeling multiple dependent categorical variables observed over time through evolving latent life stages. The primary goal is to identify the stages individuals pass through during life and to characterize distinct dynamic behavioral patterns. A key methodological innovation is that each life stage is defined by time-invariant parameters, ensuring a consistent interpretation across all time points. Unlike modern Bayesian dynamic clustering approaches, which re-estimate cluster characteristics at every time step, the strategy greatly improves computational efficiency and enables meaningful longitudinal comparisons. In addition, the model captures temporal dynamics via a time-varying partition of the study population that incorporates both abrupt structural change points and individual-level transitions between life stages. The methodology is illustrated using longitudinal data from the Italian Institute of Statistics, demonstrating its ability to reveal both individual behaviors and broader societal shifts.
