Prediction Instruments

Risk Factors

Prediction instruments used to determine the likelihood of an offender recidivating rely on two types of information or risk factors, static and dynamic. Static factors, such as criminal history, age, and family history, cannot be changed. Research consistently finds that the numbers of prior incarcerations and prior offenses, as well as certain types of nonviolent crimes, are strongly associated with (that is, they predict) the likelihood of committing future crime. As a result, most prediction instruments rely heavily on static risk factors.

Dynamic factors, although used less often in prediction instruments, are usually the targets of offender programming because they are malleable and, as such, can be changed. They include personal beliefs and attitudes, associating with others who share these beliefs and attitudes, educational level, and employment.

Prediction Methods

Predictions are only as accurate as the information on which they are based. Clinical and actuarial methods are used to gather information for predicting offender behavior. Clinical methods involve gathering information through an interview consisting of both open- and closed-ended questions with the person being assessed. The prediction is based on the interviewer's interpretation and professional judgment of the answers given, visual and verbal cues, and overall impressions of the person interviewed. Clinically based predictions are influenced by the offender's personality as well as the facts of the situation. Furthermore, accurate prediction is highly dependent on the interviewing skills of the assessor.

Actuarial methods, which have a much higher level of accuracy, may include an interview but are much more structured; specific questions and answer choices are derived from a statistical analysis of the relationships between each answer and the likelihood of the outcome being predicted. Actuarial methods shift from decision making based on professional judgment (clinical method) to decisions based on statistical relationships.

Where the information-collection process includes offender interviews, inconsistent results may occur as a consequence of poor inter-rater reliability, defined as differences in how interviews are conducted from one interviewer to the next. Problems with inter-rater reliability are more likely and a greater concern where clinical methods are employed. Interviewer training and monitoring are crucial.

Generations of Prediction Instruments

Researchers have described three generations of prediction instruments in criminal justice. The first generation involved the review of whatever information was available on the offender and then making a professional judgment. These instruments were generally informal, included a clinical interview, and were influenced by the personal opinions of the assessor. Second-generation instruments moved to structured, empirically based questions predetermined through a prior statistical analysis to be associated with the outcome. Each answer carried a numerical weight toward the final prediction score. Although much improved, these predictions were based on historical information (static factors), which limited their utility in making decisions about treatment or programming. Third-generation instruments continue to use weighted, statistically derived factors. However, in order to assess or predict current risk accurately at the time of the prediction, they use a combination of static and dynamic factors.

...