NCJ Number
97520
Date Published
1985
Length
163 pages
Annotation
This user's guide defines predictive attribute analysis (PAA), a technique for sequentially dividing groups on the basis of attributes of those groups that effectively predict a criterion; various functions and capabilities of the mainframe FORTRAN computer program for PAA are described.
Abstract
Methodological issues which have arisen about the use of the PAA as a general purpose prediction method are considered, and relate to PAA's reliability and validity. The predictive attribute analysis with validation extensions (PAAVE) computer program, an implementation of the procedure for conducting a PAA, is addressed. Attention is focused on the PAA algorithm, parameter specification, data input and manipulation, dichotimization, contingency table construction, statistical coefficients, and branch termination criteria. Additionally, analysis output and bootstrap resampling are addressed; consideration is given to processing specifications and options. Program dimensions and limitations are analyzed, and the possibility of determining up to 64 terminal subgroups, 5 PPA levels, is reported. The PAAVE program's capability of using up to seven online files for various input/output functions is noted, and the creation, initialization, and storage of these files is explained. The general use of the program is described in the context of program installation, program compilation and object-code generation, online file considerations, and program execution and output redirection. Finally, two complete PAA runs illustrate parameter specifications for both a standard analysis and for an analysis using the bootstrap validation procedures. Six references and four appendixes are provided.