Sequential Optimum Selection: A Categorical Alternative to Discriminant Analysis for Conducting Research in Assessment

Abstract

Sequential Optimum Selection, SOS, a nonparametric computer-assisted alternative to discriminant analysis, was introduced and explicated. SOS is an iterative procedure that develops ongoing decision rules for discriminating among groups, successively expunging frequency distributions or classes until all cases are accounted for. In three illustrative comparisons using clinical assessment data, SOS proved competitive with discriminant analysis in terms of over-all hit rates for parsimonious extraction of meaningful variables. Further, in one demonstration using parametric data, SOS held up better on cross-validation, While discriminant analysis will, in most cases, yield better discrimination than SOS, the latter does not make parametric assumptions and should therefore be considered a viable alternative as an exploratory procedure.