New Testing Methodology for Ordered Categorical Data

Cohen and Sackrowitz 9618715 A classical protocol for the evaluation of a treatment is to administer a placebo to a set of individuals and administer the treatment to another set of individuals. Oftentimes the response variable is an ordered categorical variable such as no improvement, partial improvement, complete recovery. The researcher is interested in answering the question: Is the treatment 'better' than the control? Many recommended testing procedures that are directed towards this question are found to be wanting. Among these are the likelihood ratio test and the Wilcoxon-Mann, Whitney test. The research involves a review of various definitions of 'better' and the development of improved testing procedures. The new procedures are computer assisted and are shown to have desirable theoretical and practical properties. The determination of whether or not a new product or new treatment or new method is better than a standard is a fundamental issue in a variety of areas including biotechnology, materials, manufacturing, and the social sciences. Current methods used to make decisions in these situations are found to be either inappropriate or, in many instances, weak in their ability to detect a superior product. The investigators study decision procedures based on a new approach which is effective in eliminating the shortcomings of previous methods. Theoretical arguments as well as extensive numerical calculations support the findings and demonstrate the magnitude of the improvements.