ABSTRACT
This chapter discusses the distinctions among causal, descriptive, and predictive uses of data. The direction in which a table should be percentaged depends upon whether the analyst wishes to describe a population or to assess the effects of an independent variable. In Unraveling the Gluecks construct prediction instruments using "factors that clearly differentiate the delinquents and nondelinquents in this research." Insofar as such predictions are concerned, the Gluecks are correct: The use of equal numbers of delinquents and nondelinquents is perfectly legitimate. The accuracy of predictions of disease from a set of symptoms depends on the proportion having the disease for the same reason that the efficiency of the Gluecks' predictive instrument depends upon the percentage of delinquents in the population on which it is used. The Gluecks' predictions for populations with different distributions of the dependent variable cannot be correct unless the data upon which the original predictive device was based were in error.
