A multicollinearity diagnostic for generalized linear models

The problem of detection of multicollinearity in generalized linear models is discussed. For this class of models the Belsley, Kuh, and Welsch (1980) multicollinearity diagnostic for the linear model is applied, performing the singular value decomposition on the scaled observed information matrix at the final solution . The performance of this adapted diagnostic in detecting collinearity is examined in detail for this class of models, in particular, the discrete response model as exemplified by the binary logistic and proportional odds regression models. The effects of centering of independent variables on the estimation of parameters and on the sensitivity of the proposed diagnostic in the presence of collinearity are also investigated.