Statistical diagnosis when basic cases are not classified with certainty

Abstract

SUMMARY A need is identified for statistical diagnostic techniques based on data sets containing cases which have not been allocated to a single diagnostic type with certainty but for which only an assessment of the probabilities of the types is available. In order to construct a statistical diagnostic system applicable to new cases it is necessary to introduce even more complex diagnostic assessments and a central part of the analysis is concerned with the reduction of such assessments to simpler, more interpretable forms. The theory provides a generalization of current statistical diagnostic techniques, and its relationship to these is discussed.