NCJ Number
99010
Date Published
1985
Length
17 pages
Annotation
The relative predictive efficiency of five parole prediction methods (Burgess, Glueck, multiple regression, association analysis, and predictive attribute analysis) was investigated using data for 854 men released on parole from the Texas Department of Corrections.
Abstract
The sample was divided randomly into a construction and validation sample. Since 142 parolees in the construction sample were successful on parole, each technique was designed to predict about 142 successes in the validation sample. In the construction sample, the more sophisticated methods (predictive attribute analysis and multiple regression analysis) were the most efficient predictors. However, in the validation sample, the efficiency ranking showed the Glueck method to be most efficient, followed by the Burgess method and multiple regression. Predictive attribute analysis dropped from first in the construction sample to fourth in the validation sample. While shrinkage was found with all methods, it was greater with the more sophisticated techniques. Overall, there was a great deal of agreement among the methods: in the validation sample, 234 subjects were accurately predicted as failures by at least 4 of the methods; 80 subjects were predicted as successes by at least 4 of the methods. The amount of variance explained by the 5 methods was almost exactly equal to the 33 percent explained when the 20 variables were used as independent variables and parole outcome was used as the dependent variable. Thus, it seems that all 5 methods combined used all of the information contained in these 20 variables. The remaining 67 percent of the variance could not be accounted for by the variables studied. Tabular data and 19 references are provided.