NCJ Number
158986
Date Published
1996
Length
36 pages
Annotation
In discussing key issues in risk prediction in criminal justice, this chapter addresses types of prediction, prediction studies in criminal justice, methodological issues, acceptable prediction accuracy, the ethics of prediction, and prediction improvement.
Abstract
The literature generally recognizes two basic forms of prediction: clinical or statistical. Clinical predictions may well be systematic but include more subjective factors in assessment. Statistical prediction is based on more objectively discernible criteria than those used in clinical prediction. Following an overview of the history of prediction studies in criminal justice, the author profiles current prediction studies in criminal justice. The discussion encompasses the prediction method used by the Edgecomb Day Treatment Center in New York City, adult probation restructuring in New York City, pretrial drug testing, pretrial guidelines, and intervention strategies for high-risk youth. A discussion of methodological issues focuses on the sample that should be used, how to measure the criterion variable, how to select and measure the predictors, what to do about missing data, how to combine the predictors, assessment of predictive power, instrument validity, and static and dynamic prediction. In discussing acceptable prediction accuracy, the author advises that with large samples, a prediction model that explains just 10 percent of variance can produce statistically significant results. A discussion of the ethics of prediction notes that the correct way to remove the effects of status -- gender, race, religion, etc. -- from prediction requires that prediction research be a two-step process that involves the independent stages of estimation and validation of a model and the implementation of a model to predict future behavior. The author's recommendations are summarized in an explanation of how to improve prediction. 23 notes and 3 tables