arXiv · 2611.18885
Multicollinearity diagnosis in generalized linear models: A systematic review and empirical reassessment
Abstract
Ridge-type estimators are widely used to address multicollinearity in generalized linear models (GLMs), yet their empirical justification depends heavily on how multicollinearity is diagnosed. Following systematic review guidelines, we conduct a citation-based review of 120 real-data applications to examine current diagnostic practices. The findings reveal a lack of consensus across studies: approaches vary regarding data transformation prior to diagnosis, applying transformations to either the design matrix or the weighted information matrix, while others omit transformation entirely. Nevertheless, the condition number emerges as the most widely used diagnostic. We review existing methodologies for computing the condition number in GLMs and recommend the optimal approach, as it uniquely attains its minimum value under orthogonality while retaining the intercept. Finally, these formulations are implemented in R and used to replicate five real-data applications from the literature.
Explore related subjects
Keep this discovery
Explore connections, maps & timelines
Catalina B. García-García, Jorge L. Bazán Guzmán, Román Salmerón-Gómez. 2026-11-13. Multicollinearity diagnosis in generalized linear models: A systematic review and empirical reassessment. https://searcharxiv.com/abs/2611.18885
Citations retain the original authors and DOI and include this site’s record URL. If this discovery service contributed to your research, you can also cite Searcharxiv by Indiciae: BibTeX · RIS.