McMaster University
Skewed Matrix-Variate Covariance Pattern Mixture Models for Longitudinal Data: A Comparison with Matrix-Variate Cluster-Weighted Models
Abstract
dc:description.abstractFinite mixture models provide a flexible framework for clustering heterogeneous longitudinal data. The CPMM and the CWM offer complementary approaches for capturing temporal and cross-variable dependencies in matrix-valued observations. The thesis develops a family of SMV-CPMM by extending the original matrix normal CPMM to four skewed matrix-variate distributions, including MVST, MVGH, MVVG, and MVNIG, constructed via generalized inverse Gaussian (GIG) variance-mean mixtures. Parameter estimation is performed using a tailored ECM algorithm. Extensive simulation studies evaluate MV-CPMM performance under varying cluster separation, sample sizes and model selection strategies. A second simulation examines the relative robustness of the MV-CPMM and MV-CWM framework under different levels of skewness. Applications to two real longitudinal datasets reveal complementary strengths. When cluster structure is primarily driven by response dynamics, CPMM yields more meaningful group recovery, whereas CWM performs better when clustering is driven by covariate variation. These findings confirm that the effectiveness of each approach depends on the underlying source of cluster separation. Overall, the proposed SMV-CPMM framework provides a flexible and robust approach for clustering complex matrix-valued data.
Degree
thesis:*- Department dc:contributor.department
- Mathematics and Statistics
- Year dc:date.issued
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Li, Naien
- Advisor dc:contributor.advisor
-
- McNicholas, Paul
Subjects
dc:subject × 6Rights
dc:rights- Statement dc:rights
-
- Attribution-NonCommercial-NoDerivs 2.5 Canada
- Licence dc:rights.uri
- Language dc:language.iso
- en
Identifiers
dc:identifier.*- Identifier URI
- https://doi.org/10.71548/43