March 5, 2010

Study: Actuarials fail to predict sexually violent recidivism

In a new prospective study out of Austria, none of the actuarial instruments commonly used to predict sex offender recidivism were able to predict sexually violent recidivism among a group of sex offenders released from prison after treatment.

The interesting study, just published in the International Journal of Offender Therapy and Comparative Criminology, was designed to validate German versions of commonly used actuarial tools, including the Static-99, RRASOR, SORAG, and SVR-20. It followed about 400 Austrian prisoners for an average of three years in the community.

The main problem obtaining significant results was that recidivism was so rare. Obviously, the less likely an event is to occur, the harder it is to accurately predict. Only seven offenders in the entire sample committed a new hands-on offense during the followup period, and most of those were extrafamilial child molesters. Recidivism base rates were especially low for rapists and incest offenders.

The results echoed findings in two other recent studies in which the actuarials failed to demonstrate good predictive validity for predicting sexually violent reoffending.

Most of the instruments did better when recidivism was defined more broadly, to include all sexual reconviction, even hands-off offenses such as voyeurism or exhibitionism that is not typically defined as sexually violent under civil commitment laws. Even including these lesser offenses, the overall base rate for all sexual recidivism among this sample was still quite low, 4.3% (12% among extrafamilial child molesters, 1.7% among rapists, and about 1% among incest offenders).

When extrafamilial child molesters -- the group most likely to reoffend -- were examined separately, all of the instruments except the RRASOR had some predictive utility, with the SVR-20 doing the best. Still, neither the Static-99 (the most widely used actuarial tool) nor the RRASOR could significantly predict sexually violent reoffenses even for that relatively higher-risk group.

"From the results of these studies and of the present study, the actuarial prediction of some reoffence categories in at least some offender subtypes is less accurate than generally assumed,” the authors concluded. "One major aim of most criminal justice systems is to calculate risk by predicting the probability of severe sexual crimes. This goal obviously is not yet achieved satisfactorily by actuarial risk assessment, because results are far from ideal, especially when time-at-risk periods are relatively short."

An important implication of this study is that evaluators need to consider offender subgroups separately, rather than lumping all types of sex offenders together. Recidivism varies tremendously by type of offender (e.g., rapists versus child molesters) and by how recidivism is defined, with the various instruments doing better at some types of predictions than others. Furthermore, so little outcome research exists on certain groups (such as hands-off offenders, juveniles, the intellectually disabled, and offenders with only adult male victims) that the actuarials may be inappropriate to use at all.

The study is:

Rettenberger, M., Matthes, A., Boer, D.P., & Eher, R. (2010).
Prospective Actuarial Risk Assessment: A Comparison of Five Risk Assessment Instruments in Different Sexual Offender Subtypes. International Journal of Offender Therapy and Comparative Criminology, 54 , 169-186.

Hat tip: Jeffrey Singer

FURTHER READING: For those of you interested in the actuarials, I also recommend "More prejudicial than probative?," a stastical critique by David J. Cooke, a forensic psychology professor in Glasgow who is an expert scholar and trainer on violence risk assessment. Cooke argues that the actuarials are compelling because they are simple to use by paraprofessionals and have a scientific veneer, but "the scientific basis for actuarial scales, as applied to individuals, may be more illusory than real." The article, in the journal of the Law Society of Scotland, is available online. It also includes useful references to other sources.


Anonymous said...

There are two parts to this study that puzzle me (based upon your reporting).

(1) I wonder about the definition of violence. There was recent Supreme Court case that decided this topic and one thing it illustrated is that "violence" means widely and wildly different things to different people. There has been no standard definition of what a sexually violent crime is, at least within the legal system. The failure might not be in the assessments themselves but the fact that they are being applied to subjects they were never intended for.

(2) Second, I'm always bothered by the short follow-up time frames. I have heard people claim that while the rate of recidivism is low in the short run in the long term (20 years) it's actually high (around 30%). I've heard people say, "Well, it's true that they won't offend again right away but they will sooner or later and that's why it's so important to keep them locked up for life!"

Karen Franklin, Ph.D. said...

In this study, sexual violence was defined based on the definitions proposed by Quinsey, Harris, Rice, and Cormier (2006): "In a nutshell, they decided to count all sexual assaults involving physical contact as violent. By contrast, sexually motivated non-contact or hands-off offences such as exhibitionism and voyeurism did not count as violent." I know others define it differently.

Regarding your second point, it was a prospective study designed to test the validity of German-language versions of the instruments, so obviously they couldn't wait around 20-30 years to see if the guys eventually reoffended. Presumably, these guys can still be followed as time keeps ticking along.

But my understanding is that most recidivism happens within the first 1-3 years, anyway.

Anonymous said...

You might be talking about a 1997 study

Prentky, Lee, Knight, and Cerce (1997) found that over a 25-year period, child molesters had higher rates of reoffense than rapists. In this study, recidivism was operationalized as a failure rate and calculated as the proportion of individuals who were rearrested using survival analysis (which takes into account the amount of time each offender has been at risk in the community). Results show that over longer periods of time, child molesters have a higher failure rate—thus, a higher rate of rearrest—than rapists (52 percent versus 39 percent over 25 years).

Anonymous said...

Anonymous copy/pasted from the site, which uses the Prentky as an example of one extreme. In the very next section and immediately following the Prentky study reference is this: "Making Sense of Contradictory Findings."

See "An Analysis of the Argument That Clinicians Under-predict Sexual Violence in Civil Commitment Cases" that critiques the problem with the Prentky study. One obvious problem is that it was a study of 115 extrafamilial imprisoned molesters "convicted of many more sex offenses (N=4.6) than a more typical sample of incarcerated sex offenders, about 90% of whom have been convicted of only one sex offense (Song & Lieb, 1995)."

Nor was the recidivism the stereotypical molestation and it included "(e.g., statutory sex offenses, public indecency, communicating with a minor for an immoral purpose)." Of course the reoffense rate would be higher than average.