Collinearity and Ridge Estimation in Generalized Linear Models

DÍAZ, MP; MUÑOZ, SE

Rio de Janeiro, Brasil

Congreso; 69th Annual Meeting of the Institute of Mathematical Statistics; 2006

IMPA, Rio de Janeiro, Brazil

Multicollinearity is an usual phenomenon found in practice when a set of variables is used to estimate regression coefficients. We study the collinearity problem of the logistic regression into generalized linear models approach. We construct an alternative tool to help the fitting model process for the case of several correlated covariates and define the deviance inflation factor (DIF) to detect this phenomenon in order to extent of some of the criteriums used in the classical regression analysis, now into the generalized linear models framework. Also, we construct SAS macros to obtain ridge estimator fot the parameter vector. We believed that this development is an usefull statistical tool to deal with collinerity in this model family.