'Model selection for generalized linear models with factor-augmented predictors' DISCUSSION

Abstract

Research work undertaken in the subject of model selection for generalized linear models with factor-augmented predictors is reviewed. The studies have found that the traditional principle component analysis fails to be mostly effective approach for dimensional reduction in regression, as the role of the response is completely ignored, and the resulting principle components might not be sufficiently predictive to the response. Computational details of inverse regression methods reveal that they are a supervised version of dimension reduction, where the effect of the response is taken into consideration. The supervised principle component analysis, a two stage procedure, is also considered in research articles. A set of relevant variables are selected by simple univariate regression, which leads to a reduced data set.