Prediction: An Overview of Selected Methodological Issues

A variety of statistical methods are available for making criminal justice predictions. None possesses a clear-cut advantage over the others, and this may be due to the poor quality of available data. Factors that affect the accuracy of predictions, regardless of the method used, include measurement reliability, base rates (the relative frequency of an event in the study population), selection ratios (the proportion of events identified by the prediction method as belonging to the outcome class), sample representativeness, and cross validation. The present ability to predict the decisions of criminal justice functionaries or offender behavior is modest. Predictions of decisions are more accurate than are predictions of behavior, and empirical studies indicate that statistical predictions outperform intuitive or clinical predictions. Predictive accuracy can be enhanced by improving data reliability, by improving measurement of prediction and outcome variables, by using both statistical and subjective bootstrapping models, by using theory-driven approaches, and, finally, by devoting greater attention to ethical implications of prediction methods and uses.