1 Developments in Risk Assessment Violence Risk and

1. Developments in Risk Assessment: Violence Risk and Sexual Violence Risk Kirk Heilbrun, Ph. D. Drexel University kirk. heilbrun@drexel. edu http: //www. drexel. edu/psychology/ research/labs/heilbrun/publications/

2. Goals • Describe risk assessment (RA) tools • Describe sexual violence RA tools • Review empirical studies on violence and offending in different populations • Describe scientifically-supported, scientifically-unsupported, and controversial uses of risk assessment

3. Important Conceptual Advances Since 1980 • • • Short vs. long-term prediction Use of term “risk assessment” Risk-Need-Responsivity (RNR) and riskneeds assessment • Particular attention to situational variables • Actuarial vs. SPJ vs. unstructured professional judgment

4. Components of “dangerousness” • Risk factors, protective factors – variables used to predict outcome • Harm (nature and severity) • Risk level – probability that harm will occur

5. Nature of risk factors • dynamic ‑ changeable via intervention with individual (treatment, monitoring) or control of situation (living setting, access to weapons) – stable – acute • static ‑ not changeable via such intervention; may include personal characteristics (age, gender) and certain kinds of disorders or deficits (psychopathy, mental retardation)

6. Legal Contexts • • • Criminal responsibility Sexually Violent Predators Capital sentencing Civil commitment Correctional transfer Workplace disability 12/5/2020 6

7. Legal Contexts • • Child custody Child protection Juvenile disposition and transfer Tarasoff 12/5/2020 7

8. Legal Standards: Risk Assessment Components • • • Nature of risk factors Level of risk Severity of harm Length of outcome period Context in which harm may occur 12/5/2020 8

9. Steps in FMHA Risk Assessment • Is violence risk part of the evaluation? • Selection of data sources • Conducting interviews, administering measures, and reviewing records • Interpretation of results • Communication of findings • Judicial decision • 12/5/2020 9

10. Forensic Mental Health Concepts • • • Context – domains, FMHA risk assessment Purpose – why conducting the evaluation Populations – with whom Parameters – structuring it Approach – procedures and specialized tools 12/5/2020 10

11. Purpose • • • Prediction/classification Management/intervention planning Both (risk-needs) 12/5/2020 11

12. Population • • • Age Gender Mental health status Location Racial/ethnic group 12/5/2020 12

13. Parameters • • • Target behavior Frequency Probability/risk category Settings Outcome period Risk and protective factors 12/5/2020 13

14. Approach • • Actuarial (predictive, risk-needs) Structured professional judgment (riskneeds) Anamnestic (needs) Unstructured clinical judgment 12/5/2020 14

15. Empirical Foundations and Limits Actuarial - formal method using equation, formula, graph or table to arrive at a probability or expected value of some outcome. Uses quantified predictor variables validated through empirical research 12/5/2020 15

16. Empirical Foundations and Limits Structured Professional Judgment – uses specified risk factors, not necessarily from one dataset. Items are carefully operationalized so their presence can be reliably coded. Evaluators then weight the presence of risk factors and anticipated intensity of management/treatment needs is drawing conclusion about risk 12/5/2020 16

17. Empirical Foundations and Limits Anamnestic – process using applied behavior analytic strategies, seeking detailed information from the individual regarding previous behavior similar to the target outcome. “Individualized” risk factors are then derived. 12/5/2020 17

18. Psychosis as Risk Factor for Violence • • Mac. Arthur Risk Study – psychosis alone not risk factor; combined with substance abuse highest risk (slightly higher than SA alone) Douglas, Guy & Hart (2009) meta-analysis —psychosis results in 49 -68% increase in odds of violence; moderated by study design, definition and measurement of psychosis, and comparison group 12/5/2020 18

19. Using Actuarial Measures with Individuals: Debate • Discussants – – – • • Hart et al. (2007) Harris & Rice (2007) Mossman (2007) Confidence intervals Wilson’s formula (N=1) 12/5/2020 19

20. Empirical Evidence on Actuarial Prediction • • Longstanding area of study (Meehl, 1954) Meta-analyses – – • Bonta, Law, & Hanson (1998) Gendreau, Goggin, & Smith (2002) – PCL-R vs. LSI-R Walters (2003) – PCL-R vs. Lifestyle Criminality Screening Form Leistico et al. (2008) – PCL-R Mac. Arthur Risk Study 12/5/2020 20

21. Empirical Evidence on SPJ • • • 13 studies (12 published, 1 dissertation) 11 suggest SPJ judgments are significantly predictive of violent recidivism 2 did not support this relationship 5/5 studies concluded that SPJ “final judgment” adds incremental predictive validity to the actuarial combination of tool elements Heilbrun, Douglas, & Yasuhara (2009) 12/5/2020 21

22. Empirical Evidence on SPJ vs. Actuarial Approaches • Limited evidence – – • • 4 studies compared approaches 2 found no differences; 2 favored SPJ in predictive accuracy Enhanced structure associated with actuarial or SPJ approaches increases accuracy (Monahan, 2008) Kroner et al. (2005) “coffee can” study— may be reaching ceiling on predictive accuracy 12/5/2020 22

23. Hanson & Morton-Bourgon sex offender meta-analysis (2009) • • • 118 samples (N=45, 398), 63% unpublished Best support for actuarial (e. g. , Static-99) and mechanical (e. g. , add scales on SVR-20) approaches Intermediate support for SPJ approaches, although SVR-20 stronger Weakest support for unstructured clinical judgment No support for “adjusted actuarial” 12/5/2020 23

24. Actuarial vs. SPJ Evidence on Predictive Efficacy • • • Depends on tool Limited evidence comparing them directly Existing evidence suggests the two approaches are comparable (Heilbrun, 2009) 12/5/2020 24

25. Singh/Fazel Metareview (2010) • • Investigated quality and consistency of findings in reviews and meta-analyses 40 reviews comprising 2, 232 studies; nine main findings Clinical, actuarial, SPJ: 5/6 meta-analyses found more support for actuarial than clinical prediction; 6 th meta-analysis found actuarial and SPJ comparable Measures: No one measure was consistently better than all others 12/5/2020 25

26. Singh/Fazel Metareview (2010) • • Country of study inconclusive—two reviews found larger effects in U. S. , a third concluded the opposite Gender: 11/13 reviews concluded that tools worked comparably in males and females Ethnicity: 5 reviews found no differences; 3 reported that greater effects resulted from higher proportion of Whites Psychiatric populations: 1 meta-analysis found larger effects in PP; 3 found no diffs 12/5/2020 26

27. Singh/Fazel Metareview (2010) • • • Definitions of outcome: included rearrest, reconviction, reincarceration, nonaggressive misconduct, general aggression, physical violence, verbal aggression, and property destruction Length of outcome period: 4/6 meta-analyses found that length of outcome period was not related to effect size Risk factors: both static and dynamic risk factors linked to repeat offending 12/5/2020 27

28. Singh/Grann/Fazel Metaregression (2011) • • Systematic review and meta-analysis using 9 risk assessment instruments (HCR-20, LSIR, PCL-R, SORAG, SVR-20, SARA, Static 99, SAVRY, and VRAG) 68 studies/25, 980 participants/88 independent samples Highest predictive validity: SAVRY Lowest predictive validity: LSI-R, PCL-R 12/5/2020 28

29. Yang/Wong/Coid Meta-analysis (2010) • • Meta-analysis using 9 risk assessment instruments (HCR-20, LSI/LSI-R, PCL-SV, VRS, OGRS, RM 2000 V, and GSIR) 28 studies published between 1999 -2008 25% of variance related to differences between tools; 85% of study heterogeneity was methodological All 9 were moderately successful and hence interchangeable except PCL-R with men 12/5/2020 29

30. Major Developments in Risk Assessment Tools • See Otto & Douglas (2009) for overview of major risk assessment tools • Prediction – – – VRAG, SORAG RRASOR, Static-99, Static-2002 COVR • Risk Reduction – Analysis of Aggressive Behavior

31. Major Developments in Risk Assessment Tools • Risk-Needs – – – – HCR-20, SVR-20 RSVP Stable 2000, Stable 2007 Acute 2000, Acute 2007 VRS, VRS-SO version LS/CMI SARA SAVRY, YLS-CMI, WAJA

32. SAVRY (Borum et al. , 2005) • Structured clinical assessment • 25 items • Items are scored -/+ – Historical items – Social/Contextual items – Individual/Clinical items – Protective items

33. SAVRY (Borum et al. , 2006) • Historical Items, e. g. , – Violence history, non-violent offense history, violence in the home, early onset of delinquent behavior, parental criminality, poor school achievement • Social/Contextual Items – Peer delinquency, peer rejection, poor parental involvement and management, lack of personal and social support

34. SAVRY (Borum et al. , 2006) • Individual/Clinical Items – Impulsivity, substance abuse, anger management problems, psychopathic traits • Protective Items – Prosocial peers, strong social support, strong school commitment, open to intervention, strong attachment to adult role model

35. YLS/CMI (Hoge & Andrews, 2002) • • Prior and Current Offenses/Dispositions Family Circumstances/Parenting Education/Employment Peer Relations

36. YLS/CMI (Hoge & Andrews, 2002) • • Substance Abuse Leisure/Recreation Personality/Behavior Attitudes/Orientation

37. Level of Service/Case Management Inventory (LS/CMI) • • • Andrews, Bonta, & Wormith (2004) Actuarial risk-needs tool, RNR influence Designed for correctional population Highly reliable (internal consistency) Predictive validity comparable to or better than PCL-R (Gendreau et al. , 2002)

38. Violence Risk Scale (VRS) • • • Actuarial risk measure, RNR-based 6 static, 20 dynamic items Male adult offenders Good reliability (ICCC >. 80) Good predictive validity (AUC=. 75 for violent reconviction, . 72 for nonviolent reconviction (Wong & Gordon, 2006)

39. Historic-Clinical-Risk Management (HCR-20) • Webster et al. (1997) • SPJ risk-needs tool, risk factors in 3 domains • Historic: largely static • Clinical: dynamic • Risk Management: dynamic • Validation research has been conducted predictively, using H domain

40. Violence Risk Appraisal Guide (VRAG) • Almost entirely historical and static factors • Derived on Canadian sample of mentally disordered offenders • Relatively long outcome period (means of 7 and 10 years, respectively) • Actuarial tool, strength is prediction

41. Classification of Violence Risk (COVR) • Based on data obtained from Mac. Arthur risk project (Monahan et al. , 2001) and additional 2 sites (Monahan, Steadman, Robbins et al. , 2005) • Chart review, brief interview, computer entry/scoring, decision tree methodology • Civil commitment, not criminal • Good reliability and validity (Monahan, Steadman, Appelbaum et al. , 2005)

42. Sexual Offender Risk Appraisal Guide (SORAG) • Quinsey et al. (2006); 14 risk factors (13 static) • Predictors of recidivism in last decade: sexual deviance, young age, offending hx, juvenile antisociality, psychopathy or personality disorder, alcohol abuse, extrafamilial victims, abused or lived apart from parents as a child

43. Static-99 • Hanson & Thornton (1999); Harris et al. (2003) • Ten items 4 risk levels • Created merging RRASOR & SACJ-Min • Highly reliable (inter-rater) • Predictive validity: AUC values around. 70 (good) (Anderson & Hanson, 2009)

44. Static-2002 • Updated version of Static-99 (Hanson & Thornton, 2003) • Limited available research suggests comparable reliability and validity to Static-99 • For use with adult males charged w/ or convicted of offense w/sexual motive • Official records needed

45. Stable 2000, Stable 2007 • Dynamic risk factors account for variance beyond static predictors (Anderson & Hanson, 2009) • Measures stable dynamic needs (contrast w/acute) for sexual offenders • From Sex Offender Needs Assessment Rating • Can be combined w/Static-99, Static-2002

46. Acute 2000, Acute 2007 • See Anderson & Hanson (2009) • Aggregated “acute” measures predicted better than recent acute measures • AUC=. 77 for Static-99; AUC=. 81 for Static-99 + Stable 2007 (Hanson et al. , 2007) • Suggests preference for stable factors and aggregated acute measures in FMHA

47. Sexual Violence Risk-20 (SVR-20) • Boer et al. (1997) • SPJ risk-needs tool; structure is somewhat similar to HCR-20 • Fewer dynamic risk factors • Three domains – Psychosocial adjustment – Sexual offenses – Future plans

48. Risk for Sexual Violence Protocol (RSVP) • SPJ tool • SVR-20 and RSVP conceptualize risk to include nature, severity, imminence, frequency, and likelihood (contrast w/actuarial) • Civil and criminal applications, males 18+ • Reliability good to excellent

49. Risk for Sexual Violence Protocol (RSVP) • 22 items in 5 domains: sexual violence hx, psychosocial adjustment, mental disorder, social adjustment, manageability • Limited validity data to date, but looks promising (Hart & Boer, 2009)

50. Violence Risk Scale-Sexual Offender version (VRS-SO) • Adapted from VRS • 7 static, 17 dynamic items • Good reliability (ICCC=. 74 -. 95) (Beyko & Wong, 2005) • Good predictive validity (static AUC=. 74, dynamic AUC=. 67, total AUC=. 72) (Wong & Olver, 2009)

51. Analysis of Aggressive Behavior • See Appendix • Individualized assessment based on “anamnestic” approach • Uses individual’s history to identify risk and protective factors • Useful for risk reduction but not prediction • Links to treatment planning

52. Research on Violent Behavior in Individuals w/Mental Disorder • Increasingly sensitive measures yield higher base rates • Particular importance of Mac. Arthur Risk Study (size, methodology, multi-site) • Importance of co-occurring substance abuse as risk factor

53. Research on Violent Behavior in Sexual Offenders • Methodological issues – Measuring outcome • • • Nature of behavior Accuracy (often underreported) Duration of outcome period

54. Base Rates of Sexual Offending • Investigators report wide range of recidivism • Underreporting is problem • “adjusted actuarial” debate • Treatment outcome data • Outcome rates for Static-99

55. Sexual Offenders: How Specialized? • Are we only concerned with sexual reoffending? • What about nonsexual offending? • Nonsexual violence?

56. Adolescent Sexual Offenders: How Specialized? • Seto & Lalumiere (2010) meta-analysis of 59 studies comparing male adolescent sexual offenders (n=3, 885) with male adolescent non-sexual offenders (n=13, 393) • Results do not support explanation of general antisocial tendencies

57. Adolescent Sexual Offenders: How Specialized? • Seto & Lalumiere (2010) found empirically supported differences in sexual abuse hx, exposure to sexual violence, other abuse or neglect, social isolation, early exposure to sex or pornography, atypical sexual interests, anxiety, and low self-esteem

58. Implications for Risk Assessment with Sexual Offenders • Need to compensate for underreporting • Need to include both stable (risk status) and changeable (risk state) elements • Static-99 or SORAG measure risk status • What assesses risk state? Stable 2007 and Acute 2007 as possibilities.

59. Risk Factors for Sexual Offending • See measures of sexual offending risk discussed to this point • Add anamnestic, individualized approach

60. Scientifically-Supported Approaches to Risk Assessment • Conclusions that individuals scoring higher on validated actuarial or SPJ tool are at greater risk for violence • Actuarial predictions for groups, given large validation samples and including margin of error

61. Scientifically-Supported Uses of Risk Assessment • Extreme risk categories as more informative • Indicating that applying group-based data to individual or small number of cases will yield wider confidence intervals

62. Scientifically Unsupported Uses • Actuarial prediction strategies without large derivation and validation samples • Actuarial prediction strategies applied to populations that are not part of derivation and validation samples • Conclusion that individual has X probability of future violence without cautions about CIs and less certainty in individual cases

63. Scientifically Controversial and/or Untested Uses • Actuarial prediction strategies with large derivation and validation samples using mean probability but not citing margin of error and greater uncertainty in individual case • Assumption that there are reliable, known probability estimates robust across samples, even at group level

64. Forms of Risk Communication • Prediction-oriented – Probability (“ 10% likelihood”) – Frequency (“ 10 individuals in 100”) – Include confidence intervals • Management-oriented • Risk-needs

65. Justifications for Importance of Risk Communication • Significant demand for risk assessment • Likely increase in this demand in the future • Link between risk assessment and decisionmaking • Enhances better-informed legal decisionmaking • Serious impact of risk-relevant decisions

66. Preface to Risk Assessment Report • Prediction, Management, and Risk-Needs • Can be adapted for own practice

67. Testimony on Risk Assessment “Will he be violent? ” – Refer to written statement – “It depends. ” – Respond in probability, not yes or no • “Is he dangerous? ” – Refer to written statement – Describe target behavior and time period – Ask to clarify meaning of “dangerous”