Abstract
This application of stepwise discriminant analysis explores a way to structure skewed-sample interview data, discusses some implications of the results and the conceptual limitations of assumptions about deviancy, and suggests a focus for future studies of the complex phenomena of drug abuse. The methodology differs from previous reports in that it uses a constant probability level, letting the computational procedure determine the number of significant variables. It also provides a method of determining the “chance” level (the amount of correct classification when there are no differences between the groups) so that the observed correct classification values may be usefully interpreted.