CHAPTER 6 - Test Selection Techniques

Abstract

This chapter presents the Wherry shrinkage formula Wherry test-selection method. Weighting large numbers of predictors (60 to 90) with samples not much larger (120 to 180) resulted in quite high multiple correlations, but upon cross-validation the prediction was usually much lower than if the single best predictor had been used alone. The test-selection method was developed for the U.S. Employment Service during its early days when its technicians, who were frequently confronted with numbers of predictors even larger than their samples, discovered that in such cases every known weighting technique broke down. Wherry's solution to the problem was the test-selection method. The method, which uses a modification of the Doolittle approach for solving simultaneous equations is frequently referred to as the Wherry–Doolittle method.