When asked to predict risk, we must be specific about what it is we are predicting. Risk of what? A violent crime? Specifically a sexual crime? A non-violent crime, or drug possession? Violation of conditions of supervision, which otherwise wouldn’t be criminal conduct? And predicting for how far in the future? And with what degree of accuracy?

Early predictions, even up to the 1980’s, were not at all accurate. They were described as “no better than flipping a coin.” The situation was so bad that the American Psychological Association acknowledged we don’t know how to do it, and psychologists should not attempt to make such predictions. But we needed to know, could this inmate be paroled? Can this patient be discharged from the psychiatric facility? So, the research began.

Many risk assessment instruments have been developed over the years. Many did not show good predictive accuracy. That is, when used in experimental analysis of large sets of offender data, were  they able to accurately predict which of those offenders had recidivated, and which had not? Some newly developed instruments showed the same level of accuracy as did existing instruments, but no improvement in the accuracy. Some instruments have undergone revisions, such as the Static 99 becoming the Static 99R, then the Static 2002, then the Static 2002R. Much research has been conducted (100's of published articles, and many professional seminars). Research focused on details that weren’t included in the original test. For example, how does risk change as the offender gets older (it gets lower); and should allowances be made for that in predicting a specific person’s level of risk.

The first efforts resulted in instruments that listed characteristics often observed in offenders who continued to re-offended. This is referred to as Structured (or Assisted) Clinical Judgment (SCJ), as opposed to unstructured clinical judgment or following that gut feeling “I’m sure he’ll offend again.” For example, evaluators were encouraged to take into consideration prior violence, prior sexual violence, prior non-violent offenses, lack of stable relationships, employment problems, substance abuse, mental illness, personality disorder, prior non-compliance, and minimization or denial of their offenses. Two of the best known instruments of this type are the HCR-20 (Historical, Clinical, Risk), and the SVR-20 (Sexual Violence Risk), each consisting of 20 items. These instruments are about 60 to 70% accurate, an improvement over the poor accuracy of unstructured clinical judgment.

The next approach developed was referred to as actuarial measures. These are developed not by thinking about what we know about offenders, but instead simply allowing a computer analysis to determine which characteristics are best at distinguishing recidivists from compliers. We learned that the two strongest predictors are, not surprisingly, clear evidence of persistent deviant sexual interests, and psychopathy (see below). Other predictors are age of the offender, the age and gender of the victim, marital status and prior conformity. Obviously, there is some overlap with the SCJ approach. Some of the predictors weren’t common sense, such as being raised by divorced parents-- even if both were involved in raising the child-- was associated with relapse. Surprisingly, we learned that denial of the offense is not a predictor of relapse. Examples of actuarial measures are the Static instruments-- currently the Static 2002R; the Sex Offender Risk Appraisal Guide or SORAG; and the Risk Matrix 2000. These instruments are accurate to about 70 to 80%.

Psychopathy is strongly associated with the risk of continued criminal conduct. It is an entrenched, refractory tendency to use and abuse others for personal gain. These characteristics often are evident from adolescence, if not earlier. Career criminals who have committed a diverse array of crimes over many years is an example. However, criminal conduct is not essential. A narcissist who manipulates others, violates commitments, shows no remorse, is impulsive and irresponsible and refuses to take responsibility for their actions also may be a psychopath. This is measured on the Hare Psychopathy Check List (PCL).

Most evaluators use more than one instrument. This is because in some circumstances, even the most accurate instrument can omit significant issues. For example, prior offenses are scored only if they appear in an official record. Self-report of unknown offenses are typically not included in the actuarial scoring. Thus, an offender like Jeffrey Dahmer, who had no known record of prior homicides, would have been predicted by some instruments as a moderate risk of re-offense even though his self-acknowledged problems made it clear he posed a very serious risk.

Some instruments can be used only with specific populations. For example, the actuarials listed above are used only with males over 18 who have committed an offense. The structured clinical judgment approach can be used with males and females over 18. There are specific models used only with male juvenile sex offenders. The PCL has versions for adults and juveniles and can be used with males and females. Some instruments can be used with offenders whose only offenses are non-contact offenses, such as child porn, voyeurism and exhibitionism, while other instruments cannot be used with them. Some instruments can be used with defendants with lower intellectual functioning, while others should not be used because there are no norms for that group.