The distribution of the ratios of characteristic roots (condition numbers) and their applications in principal component or ridge regression (Q1819862): Difference between revisions

Multicollinearity in regression analysis is indicated by the condition numbers, namely, the ratios of the latent roots of the variance- covariance matrix or the correlation matrix. This paper considers the distributional properties of these numbers (including asymptotic results) and successfully derives their distributions under fairly general assumptions. Examples illustrate the importance of the numbers for ridge regression and latent root regression.