The Shrinkage of Point Scoring Methods

Abstract

Point scoring, widely used in criminology and other social sciences, is a simple way of predicting a binary response on the basis of binary explanatory variables. Like all statistical predictors they are liable to shrinkage, working less well on a validation sample than they appear to do on the original data. The paper examines the extent of shrinkage and proposes shrinkage-adjusted predictions. The related'independence Bayes' method is also considered, and found to shrink more than the basic point scoring method. The results are applied to data from a cohort study in the development of delinquency