A Procedure for Uncovering Acceptable and Nonacceptable Mortgage Applications through Discriminant Analysis Using Ranked Data

The procedure developed in this paper uses a less biased statistical technique than conventional discriminant analysis and parallels the ranking procedure used by loan officers. A variety of univariate and multivariate statistical procedures as well as comprehensive validation methods are used to develop a "best" model. The resulting model provides more accurate classification than other studies have shown, without violating federal law regarding discrimination.