
Algorithmic Methods: NonStatistical Methods Solving Statistical Problems
Bruce Ratner, Ph.D. There are a growing number of new statistical methods, referred to as “algorithmic,” which primarily come from the fields of statistics and computer science. Algorithmic methods are nominally statistical models, or more aptly nonstatistical models, in that no effort is made to represent how the data were generated. They are nonparametric, assumptionfree procedures that let the data define the form of the model itself. Typically, the data analyst approaches a problem directly with a procedure designed specifically for that purpose. For example, the everyday statistical problems of classification (i.e., assigning class membership), and prediction of a continuous target variable are solved by the binary or polynomial logistic regression models, and ordinary leastsquares regression model, respectively. The most notable algorithmic methods for the everyday statistical problems are the decision trees (sets of ifthen rules), such as CART, CHAID, and C5.0. There is growing evidence that the number of algorithmic methods perform better than conventional statistical procedures for the tasks that algorithmic methods are designed to undertake. The purpose of this article is to introduce the GenIQ Model© as an algorithmic method, as it is a nonstatistical method solving the statistical problems of classification and prediction by uniquely unfolding the task of optimizing the decile table, or equivalently, the ROC curve, and the KS statistic. For an eyeopening preview of the 9step modeling process of GenIQ, click here. For FAQs about GenIQ, click here. 1 800 DM STAT1, or email at br@dmstat1.com. 
