Designed experiments for classification problems.

Abstract

This paper presents a new approach to evaluating classification procedures experimentally in human service organizations. Methods of design and analysis are proposed which answer the major objections to designed experiments with human subjects and which are suited to the special character of the classification problem. One important feature of the approach recommended here is the use of EVOP, a methodology originally developed for use in industrial production facilities, which allows methods of experimental design to be applied even where variation must be restricted. It is also shown that constrained mixture designs, rather than ordinary designs such as factorials, form the appropriate class of experimental designs for use in this problem. The example of assignment of prison inmates to minimum and maximum security is used to clarify the presentation and to demonstrate the practical implications of the proposed approach.