NCJ Number
97521
Date Published
1985
Length
117 pages
Annotation
This report describes Predictive Attribute Analysis (PAA), a quasi-statistical technique for the sequential subdivision of groups on the basis of characteristics of those groups that predict a criterion and discusses methodological issues that bear on its use.
Abstract
An overview of the literature on the PAA method is provided, and steps comprising PAA are identified. Advantages and disadvantages of the PAA method are cited, and theoretical and empirical issues relating to the method's validity and reliability are identified. Methodological and statistical issues are also addressed. Attention is focused on models and parameters, interaction effects, model detection by PAA, and subgroup classification. Validation methods are explored, and the cross-validation method and the bootstrap method for resampling are both analyzed. Some characteristics of the two-way contingency tables that are dichotomous in both variables are discussed, and considerations involved in the use of binary-coded information are detailed. Also, statistical measures of association and prediction that are appropriate for two-way contingency tables are addressed; ways that these different statistical measures can effect the results of a PAA are delineated. Alternatives to PAA are analyzed, and empirical studies of the PAA method are reviewed. The development and implementation of the mainframe computer program required for the design and conduct of a series of simulation studies are described. Simulation analyses are examined, and appropriate applications for PAA are identified. Finally, the choice of a PAA design is reviewed. Thirty-eight references and an appendix are included.