Advancing extended redundancy analysis via generalized estimating equations: Integrating model-based eecursive partitioning for subgroup detection

Project Summary

The Canadian Longitudinal Study on Aging (CLSA) collects thousands of variables from individuals with diverse backgrounds, capturing a wide range of measures over time. Existing statistical methods can accommodate large sets of predictors and outcomes but often assume homogeneous relationships across individuals—an assumption that rarely holds in real-world observational data. This thesis develops a new statistical method that extends an existing framework by automatically detecting and modeling meaningful subgroup differences, such as those defined by age or sex. The method will be evaluated using simulations to assess key statistical properties and applied to CLSA data to examine how interrelated social isolation measures relate to health outcomes across subgroups. Current CLSA applications rely on classical subgrouping approaches that underuse the richness of the data; in contrast, the proposed model offers new insights for both methodological development and CLSA research.