Discriminant function strategies for the kindergarten prediction of reading achievement

Abstract

The present paper evaluates the application of discriminant function analysis to predictive strategies developed for the kindergarten prediction of reading achievement by the Florida Longitudinal Project. In an attempt to optimize the predictiveness of these classification equations, several problems arose that provide important information concerning the application of discriminant analysis to early detection research. These problems highlight the importance of carefully examining (a) statistical assumptions underlying discriminant analysis; (b) the statistical nature of the data and sample employed; and (c) alternative applications of discriminant analysis when certain assumptions cannot be met. Early detection research has considerable relevance for clinical neuropsychology, provided appropriate application of predictive tools is made. The results of the present study show that careful evaluation of the application of discriminant analysis is crucial for reliable prediction. In addition, it is demonstrated that improved prediction can result from optimal use of statistical methods and criterion variables.