Generalized estimating equations for longitudinal compositional data analysis

Andrea Panarotto, University of Padova

Co-authors: Manuela Cattelan, University of Padova; Ruggero Bellio, University of Udine

Abstract: Longitudinal compositional data analysis is particularly suited for scenarios where the relative proportions of components change over time, such as in shifts in microbiome compositions or in the allocation of resources in economic systems. Capturing the dependence between successive observations requires tailored methods to handle the constrained nature of the data. In this work, we propose a novel approach to longitudinal compositional data analysis, representing the observations directly on the simplex and modeling the dependence on covariates and the longitudinal aspect through generalized estimating equations.