High-dimensional Four-Level Structured Functional Principal Component Analysis for Unbalanced Level Data with Application to Concussion Data
Abstract
We implement the Structured Functional Principal Component Analysis (SFPCA) framework of Shou et al. (2015) in a four-level nested setting with arbitrarily unbalanced group sizes at every level of the hierarchy. While the SFPCA framework can in principle accommodate deeper hierarchical structures, practical implementations remain challenging when the number of curves is large and the hierarchy is highly unbalanced. Our contributions are twofold. First, we operationalize SFPCA for four hierarchical levels with arbitrary unbalancedness across units. Second, we propose a memory-efficient computational strategy that makes the analysis of large collections of curves feasible on conventional hardware. In particular, we construct the four design-specific matrices required for the Method of Moments (MoM) estimators of the level-specific covariance operators while controlling their memory footprint, which would otherwise grow prohibitively with the number of curves. We illustrate the approach using three subsets of the NCAA-DoD Grand Alliance Concussion Assessment, Research, and Education (CARE) Consortium dataset (2014-2017).
Related
- A Network-Structured Bayesian Hierarchical Model for Sparse Mutation-Drug Response Associations: Application to Cancer Pharmacogenomics
- Adaptive Functional Clustering with Structured Dependence via Variational Inference
- Sequential Functional Structured Tucker Compression for Large Language Model Attentions
- Structured Neural SDEs for Functional Calibration
- Three-dimensional animation compression optimization based on structured processing and adaptive spatiotemporal segmentation algorithm
BibTeX
@misc{werkmann2026highdimensional,
title = {{High-dimensional Four-Level Structured Functional Principal Component Analysis for Unbalanced Level Data with Application to Concussion Data}},
author = {Verena Werkmann and Alok S. Shah and Brian D. Stemper and Jaroslaw Harezlak},
howpublished = {arXiv},
year = {2026},
month = nov,
doi = {10.48550/arxiv.2611.05682},
eprint = {2611.05682},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2611.05682},
}
RIS
TY - UNPB
TI - High-dimensional Four-Level Structured Functional Principal Component Analysis for Unbalanced Level Data with Application to Concussion Data
AU - Verena Werkmann
AU - Alok S. Shah
AU - Brian D. Stemper
AU - Jaroslaw Harezlak
PY - 2026
DA - 2026/11/03
PB - arXiv
DO - 10.48550/arxiv.2611.05682
UR - https://arxiv.org/abs/2611.05682
AB - We implement the Structured Functional Principal Component Analysis (SFPCA) framework of Shou et al. (2015) in a four-level nested setting with arbitrarily unbalanced group sizes at every level of the hierarchy. While the SFPCA framework can in principle accommodate deeper hierarchical structures, practical implementations remain challenging when the number of curves is large and the hierarchy is highly unbalanced. Our contributions are twofold. First, we operationalize SFPCA for four hierarchical levels with arbitrary unbalancedness across units. Second, we propose a memory-efficient computational strategy that makes the analysis of large collections of curves feasible on conventional hardware. In particular, we construct the four design-specific matrices required for the Method of Moments (MoM) estimators of the level-specific covariance operators while controlling their memory footprint, which would otherwise grow prohibitively with the number of curves. We illustrate the approach using three subsets of the NCAA-DoD Grand Alliance Concussion Assessment, Research, and Education (CARE) Consortium dataset (2014-2017).
ER -