A2009
Title: Functional modeling of sparse longitudinal trajectories with varying observation weights
Authors: Maryam Al Alawi - Sultan Qaboos University (Oman) [presenting]
Abstract: A Functional data framework is proposed for modeling sparse longitudinal trajectories observed through unequally weighted assessments with varying temporal importance. Unlike conventional longitudinal approaches that rely on aggregate summaries or treat observations equally over time, the proposed methodology incorporates assessment-specific weights directly into the trajectory construction process, allowing highly influential observations to contribute more strongly to the Estimation of the underlying latent stochastic dynamics. Trajectory reconstruction is performed within the Principal Analysis by Conditional Estimation (PACE) framework, followed by Functional Principal Component Analysis (FPCA) to capture the dominant modes of temporal variation. To investigate the effect of explanatory variables on trajectory evolution, a function-on-scalar regression model with time-varying coefficient functions is employed. The proposed approach provides a flexible framework for analyzing sparse longitudinal processes in which observations contribute unequally over time, while offering a richer characterization of latent temporal dynamics than conventional pointwise methods. Although motivated by weighted educational assessment data, the methodology is sufficiently general to extend to broader longitudinal settings involving heterogeneous observation structures.
