On a Stepwise Procedure for Two Population Bayes Decision Rules Using Discrete Variables

The primary difficulty in the application of Bayes decision rules based on discrete data has been the problem of deriving satisfactory estimates of the likelihoods. A measure of the potential effectiveness of a t dimensional sample space is developed, a linear function of which is asymptotically distributed as chi-square. A stepwise procedure for item selection is then presented in order to derive more stable estimates of the likelihoods through a reduction of the dimensionality of the sample space. The procedure is then illustrated with application to the problem of screening for schizophrenia.