Presentation on theme: "Challenges in Sex-Offender Re-offense Risk Assessments by Gregory DeClue, Ph.D., ABPP Sarasota, FL"— Presentation transcript:
Challenges in Sex-Offender Re-offense Risk Assessments by Gregory DeClue, Ph.D., ABPP Sarasota, FL American Psychological Association New Orleans August 12, 2006
What progress have we made since WHAT DO WE KNOW ABOUT SEX OFFENDER RISK ASSESSMENT? R. Karl Hanson Corrections Research, Department of the Solicitor General of Canada Psychology, Public Policy, and Law March-June 1998 Vol. 4, No. 1-2, 50-72
What progress have we made in assessment of dynamic factors? assessment of static factors? precision? comprehensiveness?
Denying individual liberty on the basis of community protection requires a defensible mechanism for identifying those sex offenders likely to reoffend.
Reliance on expert opinion has become routine, even when such opinions have limited accuracy.
Careful attention to the empirical literature, however, has the potential to improve risk assessments for sexual offenders.
Hanson described different approaches to risk assessment, summarizes the empirical literature on individual risk factors, and reviews recent attempts to create actuarial risk scales for sex offense recidivism.
A number of historical and highly stable predictor variables have been documented (e.g., offense history and deviant sexual preferences) …
… but the research on dynamic (changeable) risk factors has been limited.
Consequently, the research is more useful for identifying high-risk offenders than for determining when they could be safely released into the community.
There are two major types of risk factors: static (fixed) and dynamic (changeable).
Static risk factors, such as a history of childhood maladjustment or prior offenses, can indicate deviant developmental trajectories and, as such, mark long-term propensities to engage in criminal behavior.
Dynamic risk factors are those that predict recidivism, have the potential of changing, and, when changed, are associated with corresponding increases or decreases in recidivism.
For sexual offender commitment statutes, risk assessments are required to answer two central questions: (a) Initially, is the offender at sufficiently high risk to justify commitment?
… and (b) once committed, has the offender's risk level decreased sufficiently to justify release? Although these questions are closely related, the answers require the consideration of different types of risk factors.
The decision to commit can be based entirely on static factors. The release decision, however, requires knowledge of stable dynamic factors.
There is a substantial body of research on static risk factors that can be used to identify sexual offenders with a long-term propensity to reoffend.
In contrast, the research on dynamic risk factors is much less developed and allows for only the most tentative conclusions concerning changes in risk level.
The consequence is that there is much more evidence to justify committing offenders than there is for releasing them.
There was surprisingly little discussion of the validity of the risk assessment in Kansas v. Hendricks, 117 S. Ct (1997).
Hendricks admitted that he had repeatedly abused children whenever he was not confined. He explained that when he gets stressed out, he can't control the urge to molest children.... He stated that the only sure way he could keep from sexually abusing children in the future was to die... [and] that treatment is bull.
Three Plausible Approaches to Risk Assessment
All risk assessment should be informed by research, but the way to best apply the research depends on the nature of the research available.
The ideal risk assessment would be informed by extensive recidivism research with offenders identical to the case at hand.
Hanson expressed his belief that there are three plausible approaches to conducting risk assessments: guided clinical, pure actuarial, and adjusted actuarial.
In the guided clinical approach, expert evaluators consider a wide range of empirically validated risk factors and then form an overall opinion concerning the offender's recidivism risk.
In the guided clinical approach, the method for translating the identified risk factors into recidivism rates is not explicitly determined.
In contrast, the pure actuarial approach evaluates the offender on a limited set of predictors and then combines these variables using a predetermined, numerical weighting system.
The adjusted actuarial approach begins with an actuarial prediction, but expert evaluators can then adjust (or not) the actuarial prediction after considering potentially important factors that were not included in the actuarial measure.
When evaluating the quality of actuarial assessments, the following questions should be considered: (1) What is the measure's predictive accuracy? For instance, given a specific base rate and cutoff score, what is the hit rate? False positive rate? Probability of recidivism? …
(2) Do independent raters obtain the same score for the same individual (rater reliability)?
(3) If the scale was developed on a different population, to what extent would it be expected to apply to the current case?
4. Are there important risk factors that have been neglected?
Hanson (1998): The major vulnerability of the current actuarial scales is their lack of comprehensiveness.
1998: Each approach has demonstrated roughly equivalent (moderate) predictive accuracy. In particular, each of these approaches can be expected to reliably identify a small subgroup of offenders with an enduring propensity to sexually reoffend.
The rate at which this highest risk subgroup actually re-offends with another sexual offense could conservatively be estimated at 50% and could reasonably be estimated at 70% to 80%.
The available research, to date, cannot identify individuals who will certainly reoffend (100% probability). Any intervention aimed at high-risk offenders will inevitably target some offenders (false positives) who could have been managed with less intensive interventions.
Special handling of high-risk offenders can, nonetheless, be justified by balancing the benefit of protection of society from the truly dangerous offenders against the cost of the measures to both society and individual offenders.
Research on static risk factors for sexual offenders has developed to the point that the best predictors are being combined into actuarial scales.
In contrast, a reliable set of dynamic risk factors has yet to be established.
There are no actuarial scales for measuring treatment outcome, and evaluators wishing to use empirically guided clinical assessments are left with very little guidance from the research literature.
Until the necessary research yields results, we can only hope that today's evaluators of treatment outcome are more accurate than their predecessors.
PROGRESS IN ASSESSMENT OF DYNAMIC FACTORS
Hanson, R. K., & Harris, A. The Sex Offender Needs Assessment Rating (SONAR): A method for measuring change in risk levels Downloaded 8/6/06 from sppcc.gc.ca/publications/corrections /200001b_e.asp. sppcc.gc.ca/publications/corrections /200001b_e.asp
If a person is considered to be high-risk to sexually re-offend, and is civilly committed and treated, how do we know if/when he is no longer likely to sexually re-offend if released?
Marques, J. K., Nelson, C., Alarcon, J.-M., & Day, D.M. (2000). Preventing relapse in sex offenders: What we learned from SOTEPs experimental treatment program. In D. R. Laws, S. M. Hudson, & T. Ward (Eds.), Remaking Relapse Prevention with Sex Offenders: A Sourcebook. Thousand Oaks, CA: Sage.
DeClue, G. (2002). Remaking relapse prevention with sex offenders: A source book, and Practice standards and guidelines for members of the Association for the Treatment of Sexual Abusers (ATSA). Journal of Psychiatry & Law, 30, (book review)
Evaluating Community Sex Offender Treatment Programs: A 12-Year Follow-Up of 724 Offenders. By Hanson, R. Karl; Bloom, Ian; Stephenson, Marylee Canadian Journal of Behavioural Science. 36(2), Apr 2004,
After an average 12-year follow-up period, no differences were observed in the rates of sexual (21.1% vs 21.8%), violent (42.9% vs. 44.5%) or general (any) recidivism (56.6% vs 60.4%) for treated and untreated groups, respectively.
Based on the complete follow-up period, the unadjusted sexual recidivism rates were 21.1% (85 out of 403) for offenders who received treatment compared to 21.8% (70 out of 321) for the comparison group.
The study does not allow conclusions about what was effective or ineffective in the CSOP interventions. The findings do suggest, however, that some highly plausible interventions may have little overall effect.
Those inclined to believe that treatment is effective will emphasize that the interventions examined in the current study did not meet contemporary standards;
those inclined to doubt the effectiveness of treatment will emphasize the methodological weaknesses in the other studies that have shown positive treatment effects.
Friedrich Lösel Director of the Institute of Criminology at Cambridge
Overall, we have not found that more recent programs are superior in outcome. Although treatment before the 1970s was clearly ineffective, neither programs from the 1990s nor publications after 2000 reveal stronger effects than in previous decades. …
Even within the cognitive-behavioral category, more current programs are not more effective than older ones. …
Lalumière, M. L., Harris, G. T., Quinsey, V. L., and Rice, M. E. (2005), p. 129 Conclusion RE Sex-Offender Treatment:
Because evidence of treatment efficacy is so weak, careful risk appraisal must be a cornerstone of the management of sex offenders. Because there is scant evidence that a high-risk rapist can be turned into a low-risk offender by participation in any treatment …
The best advice is to use treatments that (a) fit with what is known about rapists specifically and about offenders and sex offenders more generally,
(b) Have been shown to produce pre- and post-treatment changes in empirically relevant measures,
(c) Are acceptable to offenders and ethically supportable,
(d) Are carefully described so that program integrity can be evaluated, and (e) can be incorporated into ongoing supervision program. …
Community supervision programs (and empirically based ways to deny access to the community) are especially important given the current lack of knowledge about effective treatments.
Risk Factors for Sexual Recidivism:
Hanson, R. K. & Morton-Bourgon, K. (2004). Predictors of sexual recidivism: An updated meta- analysis, Downloaded 2/28/06 from sppcc.gc.ca/publications/corrections /pdf/200402_e.pdf.http://ww2.psepc- sppcc.gc.ca/publications/corrections /pdf/200402_e.pdf
The Characteristics of Persistent Sexual Offenders: A Meta- Analysis of Recidivism Studies. By Hanson, R. Karl; Morton- Bourgon, Kelly E. Journal of Consulting and Clinical Psychology. 73(6), Dec 2005,
Research to watch out for:
Hart, S., D., Michie, C., & Cooke, D. J. (in press). The precision of actuarial risk assessment instruments: Evaluating the "margins of error" of group versus individual predictions of violence. British Journal of Psychiatry.
Conclusions: ARAIs cannot be used to estimate an individuals risk for future violence with any reasonable degree of certainty and should be used with great caution or not at all. In theory, reasonably precise individual estimates could be made using ARAIs if developers used very large construction samples and if the tests included few score categories with extreme risk estimates.
Evaluating Sexual Offenders: Actuarial Assessment and Beyond David Thornton, Ph.D. Sand Ridge Secure Treatment Center
The Problem There are a range of actuarial instruments with similar predictive accuracy but which give conflicting assessments of particular individuals Solutions –Pick the best scale (but which is best varies between samples) –Average –Only weight consistent results –Personal favorite –Dimensional integration
Dimensional Integration Decompose scales into underlying dimensions Predictive value carried by underlying dimensions rather than by particular operationalizations of them Understand discrepancies between scales in terms of –Weighting of particular dimensions –How scale operationalizes dimensions
Knight and Thornton Variables from Static-99, Static-2002, RM2000, MnSOST-R, VRAG, SORAG, ASOAP, SVR-20 Included static historical variables, criminal history, childhood and adolescent crime related behavior, simple demographics Excluded abstract trait variables, needs and vulnerabilities N=458 cases where no missing data
Factor I: Persistent General Criminality VRAG Non-violent RM Criminal Apps SVR Past non-sexual, non-violent offenses S02 Sentencing Apps S99 Sentencing Apps ASO Multiple types of offenses ASO Adult ASB VRAG Failure on CR S02 Breach of CR ASO Juvenile ASB SVR Supervision Failure SVR Past Non-sexual violent offenses RM Violent Apps MNS Adolescent ASB ASO Ever charged before age 16 S99 Prior NSV S02 Prior NSV S02 Years Free pre Index
Factor II: Persistence and Rate of Sexual Offending S02 Prior Sexual App RM Sex App S99 Sex Priors ASO Sex Priors MNS Sex Convictions S02 High Rate of Sexual Offending SORAG Prior Sex Convictions ASO N of Sexual Abuse Victims S02 Non contact convictions RM Non contact convictions ASO Duration of Sex Offense History
Factor III: Young and Single RM-S Age on Release RM-V Age on Release S02 Age on Release VRAG Age at Index MNS Age at Release S99 Young S99 Single RM Single VRAG Marital Status
Factor IV: Violent Stranger Assault S02 Stranger Victims RM Stranger Victims S99 Stranger Victims MNS Stranger Victims ASO Expressive Aggression in Sex Offenses SVR Physical Harm to Sex Offense Victim MNS Force/Threat in Sex Offense SVR Use of Weapons or Threats of Death in Sex Offense S99 Index Non Sexual Violence VRAG Victim Injury (-)
Factor V: Male Victim Choice RM Male Victim of Sex Offense S99 Any Male Victims S02 Male Victims ASO Male Child Victims VRAG Index offense victim female (-)
Comment Scales differ in their profiles over the factors, sometimes in quite subtle ways It is more than just greater or lesser weighting of antisociality or sexual deviance The mere presence of an item in a scale does not mean that it will have a substantial influence on the total scale score How much variance in scale scores is accounted for by the five components?
Percent of Scale Variance Accounted for by the Five Components SORAG83%RRASOR78% VRAG77%MnSOSTR66% ASOAP62%RM-S73% Static %RM-V67% Static-9985%RM-C78%
Comment Two-thirds to four fifths of scale variance is accounted for by the rotated components Note that we did not have PPG for SORAG So is the remaining variance carrying useful predictive value or is all the predictive value carried by the components?
Example Static-99 –Has no significant predictive value beyond the components –The components have significant predictive value after controlling Static-99 –Static-99 underweights Sexual Persistence and Male Victims
Comment It really does seem to be the case that the five components capture all the predictive value of the simple static items (and age) in the actuarials
Factor I: Persistent General Criminality Factor II: Persistence and Rate of Sexual Offending Factor III: Young and Single Factor IV: Violent Stranger Assault Factor V: Male Victim Choice
Currently, Hanson & Morton- Bourgon (2004) meta-analysis is best data source regarding risk factors for sexual recidivism.
Research regarding static risk factors continues to progress … but we should exercise extreme caution in risk communication.
Research regarding dynamic risk factors remains in its infancy or prenatal stage.
Sex-offender risk assessment should be guided by research, but we must recognize and communicate the limitations of current scientific knowledge regarding sex-offender risk assessment.