ABSTRACT
In the previous chapters we have placed some emphasis on the distinction between responses and explanatory variables, this distinction dictating the questions judged relevant as well as the kinds of formal representation and analysis to be used. We now discuss studies in which, for important practical reasons connected with convenient collection of data, the roles of response and explanatory variables appear to be reversed. The general idea is widely applicable. The two most important areas of application are probably epidemiology, in the context of case-control studies, and econometrics, in the context of choice-based sampling. In Section 4.4 we discuss the somewhat related issue of the connection between logistic regression and discriminant analysis.
