![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() ![]() |
|
|||||||||||
![]() |
||||||||||||
![]() |
||||||||||||
You have accessed an archived page on the Public Safety and Emergency Preparedness Canada website. This material may be outdated. Please consult our new site for up-to-date information. |
![]() ![]() ![]() |
![]() |
![]() ![]()
Notes: * p < .05; **p < .01; CI = 95% Confidence Interval. As the PRA and the SRA-GA were significantly correlated (r = .57, p < .01), we examined the relative predictive utility of combining the two instruments through multiple regression analysis. Multiple regression analysis allows one to evaluate the power of combining multiple measures (in our case, the PRA and SRA-GA) to predict outcome (i.e., violent recidivism). The results of the multiple regression analyses indicated that although the SRA-GA added to the prediction of violent recidivism (R2change = .03; p < .01), the increase in predictive power of combining the two instruments was quite small (RPRA & SRA-GA = .34, p < .01) in comparison to the predictive power of each instrument alone (rPRA = .30, p < .01 and rSRA-GA =.30, p < .01). Can We Improve Upon the Risk Assessment for Violent Offenders?As noted earlier, the notion of a specialized assessment for specific sub-groups of offenders gave rise to a number of risk/need instruments tailored to predict specific types of criminal activity. The SRA‑GA is one such specialized instrument designed to predict violent recidivism. There is a debate in the literature, particularly in regards to predicting violence, whether or not specialized violence risk/need instruments predict better than general risk/need instruments (Gendreau, Goggin, & Smith, 2002). The present results, which demonstrated that the PRA, a general risk/need instrument, performed just as well as the specialized SRA-GA instrument suggests that a specialized instrument in this case added little to the prediction of violent re-offending. Nonetheless, it may be possible to improve assessments of violent recidivism in one of two ways. First, we can try to improve upon the SRA-GA by modifying the items that comprise the scale. The second approach is to integrate the best items from of the SRA-GA into the PRA. We describe these two approaches next. Improving the SRA-GA through item modification. We undertook an analysis to explore the relationship between the individual items of the SRA-GA and violent recidivism. The goal with this item analysis was to evaluate each item’s contribution to the prediction of violent recidivism. Given that the patterns of correlations for males and females were almost identical, the subsequent analyses combined all offenders. For each of the 11 items of the SRA-GA, the correlations with violent recidivism were calculated. In addition, we examined the recidivism rates associated with each possible score that an item received. Each item was evaluated based on: a) the magnitude of its relationship to violent recidivism, and b) the observed violent recidivism rates for each possible score it could receive. An item was deemed adequate when it showed a positive correlation to violent recidivism of .10 or higher and demonstrated an increase in violent recidivism across the item’s ratings with a difference of 10% or more between the highest and lowest item score. The left side of Table 2 presents the correlations of the historical items with violent recidivism and also the recidivism rates for each score on the item. The correlations for the dynamic items with violent recidivism and their associated recidivism rates are presented in the right side of Table 2. Only three items of the SRA-GA were not significantly associated with violent recidivism, specifically, Current Conviction, ‘No Contact’ Condition Violations and Awareness of Warning Signs. The remaining items were significantly related (p < .05 or p < .01) to violent recidivism with correlations ranging from r = .10 to r = .25. In addition, the items Current Conviction and ‘No Contact’ Condition Violations did not demonstrate a step-wise increase in violent recidivism rates across the three scores. Although probationers scoring 0 on the item ‘No Contact’ Condition Violations demonstrated a lower violent recidivism rate (22.8%) than those scoring 4 (31.1%), the probationers scoring 2 actually showed the highest violent recidivism rate (32.3%). Conceptually, this item attempts to assess a willingness to violate non-association orders, a factor found in previous research to be predictive of violent recidivism (Harris, Rice, & Quinsey, 1993). These results suggest that this item could be revised to simply assess the presence or absence of any known violation of a ‘No Contact’ order, regardless whether the individual is arrested. However, when we dichotomized the scoring of this item, the correlation with violent recidivism increased only slightly (r = .08, ns). Table 2. Pearson Correlations and Violent Recidivism Rates for the SRA-GA Items (N = 444)
Notes: * p < .05; ** p < .01. Finally, although Use of Weapon demonstrated a significant correlation with future violent offending, the discrimination in rates for the different scores was poor (e.g., probationers scoring 1 had a rate of 33.9% while those with the higher score of 2 had a lower violent recidivism rate of 29.4%). Once again, simplifying the scoring for this item by collapsing the categories of the threat of use with the actual use of a weapon into one category, thus giving a 0-1 scoring scheme, yielded only a small improvement in predictive accuracy (r = .12, p < .05). In summary, the item analysis suggested the following revisions to the SRA-GA. First, eliminate the item Current Conviction from the instrument as this item demonstrated little predictive or discriminative validity. Second, revise the scoring of two items. For the item Use of Weapon, previous scores of 1 and 2 (indicating a use of a weapon for threatening or to harm) could be collapsed into one score of 2. A score of 0 would remain unchanged (i.e., no weapons was used for threats or for harm). For the item ‘No Contact’ Condition Violations, scores of 2 and 4 (indicating the individual violated a ‘No Contact’ order with or without an arrest) could be collapsed into a single score of 2. In order to evaluate the predictive and discriminative accuracy of
these revisions to the SRA-GA, a new score (Revised SRA-GA) was calculated.
This Revised SRA-GA was still significantly related to the PRA (r = .59,
p < .01). Its predictive accuracy with respect to any recidivism (r
= .38, p < .01) and violent recidivism (r = .31, p <
.01) however showed almost no improvement compared to the original SRA-GA (r’s
of .38 and .30 respectively, p < .01). Finally, a multiple regression
analysis was conducted using the PRA and the Revised SRA-GA to predict violent
recidivism. The results indicated that the Revised SRA-GA significantly added
to the prediction of violent recidivism (R2change
= .03; p < .01). However, there was no improvement in
overall predictive accuracy for the revised Enhancing the PRA. In this investigation, the PRA predicted violent recidivism as well as the SRA‑GA despite the fact that the PRA does not contain any items that specifically refers to violent behaviour. Therefore, one potential way of improving the PRA is to add a few items that assess violence. Consequently, we conducted an analysis of the SRA-GA that was intended to identify items predicting violence that could be appended to the PRA. First, a multiple regression analysis examined all the items of the SRA-GA and selected the best predictors. A step-wise method was used to ensure that any collinearity would be accounted for amongst the items. This analysis identified two significant predictors: Prior Assault Convictions (b = .231; t = 5.08; p < .01) and Motivation for Treatment (b = .197; t = 4.34; p < .01). A third item, History of Aggressive Behaviour, approached significance (b = .113; t = 1.92; p = .056). Taken together, these three items were significantly related to violent recidivism (R = .33, p < .01). The item Prior Assault Convictions evaluates prior assaults that resulted in official convictions whereas the item History of Aggressive Behaviour evaluates prior assaults (or threats) that did not result in formal convictions. Next, the sum of these three items was added to the total PRA score and the predictive accuracy of this Enhanced PRA (original PRA plus the three items) was evaluated. Table 3 shows both the Pearson correlations and the areas under Receiver Operating Characteristic curve (AUC). A Receiver Operating Characteristic analysis is unaffected by base rates and selection ratios. An AUC of 1.0 represents perfect prediction whereas an AUC of .50 represents chance. Both types of statistical analyses showed the Enhanced PRA related to violent recidivism as well as any recidivism. However, as the overlapping confidence intervals indicate, the Enhanced PRA did not perform better then the original PRA or the SRA-GA. Table 3. Predictive Validity Estimates of the PRA, SRA-GA and the Enhanced PRA
Notes: All predictive validity estimates significant (p < .01); CI = 95% Confidence Interval. DiscussionThere are two main purposes for conducting a specialized risk/need assessment. One, the specialized risk/need assessment will improve prediction of a specific type of re-offending above and beyond a general re-offending risk/need instrument. Two, the specialized risk/need assessment will enhance case management decisions by identifying appropriate treatment needs/targets and assigning appropriate levels of supervision and treatment. In the case of the SRA-GA, the focus is on generally assaultive offenders and violent re-offending. Contrary to the belief that general risk/need assessment instruments are not well suited to the prediction of violence, the results from the first study indicated that the PRA does in fact predict violent re-offending just as well as the SRA-GA, a specialized tool for violence. A detailed analysis of the SRA-GA found that one item could be deleted because it showed no association with violent recidivism and that the scoring for two other items could be simplified without diminishing the predictive validity of the instrument. However, there was no appreciable improvement to the Revised SRA-GA’s predictive accuracy. Although scores on the PRA and SRA-GA were equally predictive of violent recidivism, a question was asked whether combining the two instruments would enhance the prediction of violent recidivism. Multiple regression analyses showed that when the results from the PRA and SRA-GA (or its Revised version) were combined, there was a significant, but relatively minor, improvement in the prediction of violent recidivism. Finally, one of the weaknesses of the PRA (i.e., no items directly related to violence) was addressed by identifying three items from the SRA-GA that could be added to the PRA. However, total scores on the Enhanced PRA predicted no better the original PRA.
Study 2: Evaluation of the SRA-PAMethodParticipantsA total of 613 probationers (502 males and 111 females) served as the basis for analysis. The probationers were assessed prior to 2000 (October 1996 to December 1999) to ensure a minimum two years of follow-up. Offenders were assessed with the PRA and the SRA-PA (as per policy
guidelines) because they had: (a) a current partner assault conviction Risk Assessment InstrumentsPrimary Risk Assessment (PRA). The mean PRA score was 9.6 (SD = 3.6) with 13.2% (n = 81) in the Low Risk range, 58.6% (n = 359) in the Medium Risk range, and 28.2% (n = 173) in the High Risk range. Males (M = 9.7; SD = 3.6) scored significantly higher on the PRA (t = 2.23, p < .05) than females (M = 8.9; SD = 3.4). SRA-PA. The SRA-PA contains 12 items. The first six items make up the Historic Risk Factors section: Current Convictions, Prior Partner Abuse Convictions, History of Aggressive Behaviour, Use of Weapons, ‘No Contact’ Condition Violations, and Suicide Thoughts/Attempts. All of these items are scored as 0, 2 or 4 except Use of Weapons, which is scored 0, 1, or 2. The remaining 6 items make up the Risk Factors That Change section: Acceptance of Responsibility, Victim Empathy, Attitudes Towards Violence, Awareness of Warning Signs, Relapse Prevention Skills, and Motivation for Treatment. These items are scored as 0, 1, or 2. The total SRA-PA score is the sum of all the items and scores can range from 0 to 34. Scores 10 or lower are considered Low Risk for re-offending with a violent domestic offence, scores 11 to 20 are considered Medium Risk, and scores 21 or greater are considered High Risk. The mean SRA-PA score was 14.9 (SD = 5.5) with 21% (n = 129) of the offenders classified Low Risk, 61% (n = 376) Medium Risk, and 18% (n = 108) High Risk. There was no significant difference on SRA-PA scores (t = 1.74, ns) between males (M = 15.1; SD = 5.6) and females (M = 14.1; SD = 4.8). The PRA and SRA-PA were significantly correlated (r = .60, p < .01). Measurement of RecidivismRecidivism information was coded from CPIC Records and detailed domestic violence recidivism information was gathered through Manitoba’s Computerized Offender Management System (COMS). CPIC records were received on October 30, 2002 from the RCMP. COMS was accessed in the summer of 2003 and information dated on or before October 30, 2002 was recorded. All offenders had a minimum follow-up period of 2 years from the date of assessment. General recidivism was defined as any new conviction (including technical violations) within two years of assessment date. The overall general recidivism rate was 36.1% with no significant difference between males (37.5%) and females (29.7%). Violent recidivism was defined as any
violent conviction (e.g., assault, threats, robbery, sexual assault, and weapon
offences) within two years of the assessment date regardless whether it was
domestically related or not. The overall violent recidivism rate was 15.0%,
with males having a significantly higher rate (16.9%) than females (6.3%). No
significant differences in violent recidivism rates were found between
offenders with a partner abuse index offence Domestic violence recidivism was defined as any domestic violence related arrest (e.g., assault, threatening, criminal harassment, breach of non-contact/non-association) within two years of the assessment date. Arrest was chosen rather than conviction to increase the base rate. However, even with arrest as our outcome criterion, the base rates of domestic violence remained relatively low. The overall domestic violence recidivism rate was 11.4%, with males having a significantly higher rate (13.1%) than females (3.6%). No significant differences in domestic violence recidivism rates were found between offenders with a partner abuse index offence (n = 542; domestic violence recidivism = 12.0%) and offenders with a non-partner abuse index offence but with a domestic violence history or a concern to staff (n = 71; domestic violence recidivism = 7.0%). Assessing Predictive ValidityTo evaluate the predictive accuracy of the PRA and the SRA-PA,
Pearson correlations between the total scores for the two instruments and
general, violent and domestic violence recidivism were calculated. The results
are presented in Table 4. In addition, with respect to the prediction of
domestic violence recidivism the AUC for the PRA was .62 (CI = .55 - .68) and
.61 for the The three risk levels for each instrument failed to show an orderly, step-wise progression in domestic violence recidivism rates that was statistically significant. Statistical significance in domestic violence rates was only observed at the extreme risk levels. Offenders assessed as High Risk on the PRA had a significantly higher domestic violence recidivism rate (17.6%) than male offenders assessed as Low Risk (6.3%) and offenders assessed as High Risk on the SRA-PA had a higher domestic violence recidivism rate (20.8%) than those assessed as Low or Medium Risk (9.3% and 12.1%). We further examined the relative predictive utility of combining the two instruments through multiple regression analysis (male offenders only). In this analysis, we evaluated the power of combining the PRA and the SRA-PA to predict domestic violence recidivism. The results of the multiple regression analyses indicated that the SRA-PA did not significantly add to the PRA in the prediction of domestic violence recidivism (R2change = .01; p > .05). Item Analysis of the SRA-PAAlthough the total score on the SRA-PA was significantly related to
domestic violence recidivism, the predictive validity estimates were weak and
not significantly greater than those achieved by the PRA alone. An item
analysis was conducted using the same methodology as in Study 1 but with
domestic violence recidivism as the criterion. The analysis revealed that only
one item, ‘No Contact’ Condition Violations, showed a statistically significant
association Table 4. Pearson Correlations of the PRA and SRA-PA with General, Violent And Domestic Violent Recidivism and Domestic Violent Recidivism Rates by Risk Level
Notes: * p < .05; ** p < .01; CI = 95% Confidence Interval; N/A = non-applicable due to small cell size. DiscussionThe results from Study 2 provide weak empirical support for the use of the SRA-PA to identify partner-abusing offenders of varying risk levels. Although the SRA-PA was significantly related to domestic violence recidivism, the relationship was small, and no different than that of the PRA. Additionally, the SRA-PA failed to show an orderly, statistically significant step-wise increase in domestic recidivism across risk levels. Only offenders in the High Risk range of the SRA-PA demonstrated higher domestic violence recidivism than offenders scoring in the Low and Medium Risk ranges. Furthermore, the item analysis found that only one of the 12 items of the SRA-PA was predictive of domestic violence recidivism. Overall, these results suggest that future efforts be directed towards alternative measures to assess domestic violence risk in partner abusing offenders. There are some possible measures such as the SARA (Kropp, Hart, Webster, & Eaves, 1999) and the Ontario Domestic Assault Risk Assessment (ODARA; Hilton, Harris, Rice, Lang, & Cormier, in press). However, these instruments are relatively new and also require further validation. Another option may be to develop a new risk/need assessment instrument for domestic violence cases that would incorporate the present findings and new research that was unavailable when the original SRA-PA was developed. Regardless of the option, to implement a valid and useful risk/need assessment tool for partner violence will require time and resources.
General ConclusionsEffective case management protects society by providing services to offenders that reduce the probabilities of recidivism. For each offender, decisions must be made in regards to the appropriate level of supervision and rehabilitative programs. Accurate and valid assessments of offenders’ risk to re-offend and criminogenic needs provide crucial information for these decisions. Although there are validated risk-need instruments that assess general offending, specialized risk/need tools are often used with violent offenders because it is believed that these specialized measures provide information that more accurately predict violent re-offending. The present investigation examined the predictive validity and utility of two specialized risk/need instruments: the SRA-GA designed to be used with generally assaultive offenders and the SRA-PA for partner-abusers. In the first study, we found that both the general PRA and the specialized SRA-GA were moderate predictors of violent re-offending. However, neither measure was superior to the other. Even combining the two measures did not significantly improve the prediction of violent re-offending. Item analyses of the SRA-GA suggested eliminating one item and simplifying the scoring on two other items, but these revisions resulted in no significant increase in the instrument’s predictive validity. Finally, we considered integrating three items from the SRA-GA with the PRA. Once again, we did not find higher predictive accuracy for the Enhanced PRA over the original PRA. The second study examined the PRA and the SRA-PA with partner-abusing probationers. As with the first study, both measures were equally predictive of domestic violent re-offending. However, scores on both instruments were weak predictors. Only one of the 12 items of the SRA-PA predicted domestic violence recidivism. Overall, the results indicated that the SRA-PA added little, with or without the PRA, to discriminate groups of offenders with varying rates of domestic violence re-offending. Thus, we concluded that an alternative instrument is needed to more accurately determine the risk and needs of partner-abusing offenders. In conclusion, the general risk-need instrument performed as well as two specialized measures of violent and domestic violence re-offending. The results from the two studies, however, do not necessarily mean that the development of specialized offender risk scales is unlikely to improve upon more generalized assessment instruments. What may be important is the type of behaviour that is being predicted. The determinants of general violent and assaultive behaviour may be no different than the predictors of non-violent law violations. There is some evidence that the individual predictors of general recidivism are the same as those of violent recidivism (e.g., Bonta, Law & Hanson, 1998). When these individual items are brought together to form generalized risk scales, they predict both general and violent recidivism (e.g., Gendreau, Goggin & Smith, 2002). In this study, the original PRA (AUC = .69, CI = .64 - .75) performed as well as the VRAG, a widely used scale specifically designed to predict violence (AUC = .72; http://www.mhcp-research.com/ragpage.htm). The question arises whether more specific types of violent offending require more specialized tools. An obvious example is sexual re-offending. However, even here there is the suggestion that such specialized assessments may add little in terms of prediction (Hanson & Morton-Bourgon, 2004). In study 2, the SRA-PA was a very modest predictor of domestic violence. But, the PRA also did not perform particularly well. However, another study using a general offender risk-need scale was found to predict domestic violence (Hanson & Wallace-Capretta, 2000). In the area of risk prediction for partner assault, there is astonishingly little in scale development (Dutton & Kropp, 2000). There are a few predictive validity studies of the SARA but they have been plagued by small sample sizes (Grann & Wedin, 2002). A recent study of the ODARA appears particularly promising. Hilton and her colleagues (in press) reported for the ODARA an AUC of .72 on their cross-validation sample. In general, validating partner assault scales are particularly difficult because police do not usually note on official criminal records whether the violent offence was domestic-related or not. Thus, considerable effort is required to gather the information from other sources. Finally, one of the lessons learned from the research described in this report is that we cannot take for granted a new risk instrument that was developed with the best intentions and the best expert advice available at the time. As we found in these two studies, the need for empirical evaluation cannot be underestimated. For any organization, development and on-going evaluation are prerequisites to best practices.
ReferencesAndrews, D. A., & Bonta, J. (2003). The Psychology of Criminal Conduct. 3rd Edition. Anderson Publishing Co., Cincinnati, OH. Baird, S. C., Heinz, R. C., & Bemus, B. J. (1979). Project Report #14: A two-year follow-up. Madison, WI: Department of Health and Social Sciences, Case Classification/Staff Deployment Project, Bureau of Community Corrections. Barkwell, L. (1991). Offender classification and risk assessment. Winnipeg: Community and Youth Corrections. Bonta, J., Law, M., & Hanson, K. (1998). The prediction of criminal and violent recidivism among mentally disordered offenders: A meta-analysis. Psychological Bulletin, 123, 123-142. Bonta, J., Parkinson, R., Pang, B., Barkwell, L. & Wallace-Capretta, S. (1994) The Revised Manitoba Classification System for Probationers. Unpublished report. Ottawa: Department of Solicitor General Canada. Dutton, D. G., & Kropp, P. R. (2000). A review of domestic violence risk instruments. Trauma, Violence, and Abuse, 1, 171-181. Gendreau, P., Goggin, C., & Smith, P. (2002). Is the PCL-R really the "unparalleled" measure of offender risk? A lesson in knowledge cumulation. Criminal Justice and Behavior, 29, 397‑426. Gendreau, P., Little, T., & Goggin, C. (1996). A meta-analysis of the predictors of adult offender recidivism: What works! Criminology, 34, 575-607. Glover, A. J., Nicholson, D. E., Hemmati, T. Bernfeld, G. A., & Quinsey, V. L. (2002). A comparison of predictors of general and violent recidivism among high risk federal offenders. Criminal Justice and Behavior, 29, 235-249. Grann. M. & Wedin, I. (2002). Risk factors for recidivism among spousal assault and spousal homicide offenders. Psychology, Crime, and Law, 8, 5-23. Grove, W. M., Zald, D. H., Lebow, B. S., Snitz B. E., & Nelson, C. (2000). Clinical versus mechanical prediction: A meta-analysis. Psychological Assessment, 12, 19-30. Hanson, R. K. (2002). Evaluation of Manitoba’s Secondary Risk Assessment. Unpublished Report. Solicitor General Canada. Hanson, R. K., & Morton-Bourgon, K. (2004). Predictors of sexual recidivism: An updated meta-analysis. (User Report 2004-02). Ottawa: Public Safety and Emergency Preparedness Canada. Hanson, R. K., Morton, K. E., & Harris, A. J. R. (2003). Sexual offender recidivism risk: What we know and what we need to know. In R. A. Prentky, E. S. Janus, & M. C. Seto (Eds.), Sexually coercive behavior: Understanding and management (pp. 154-166). Annals of the New York Academy of Sciences, Volume 989. Hanson, R. K., & Thorton, D. (2000). Improving actuarial risk assessments for sex offenders: A comparison of three actuarial risk scales. Law and Human Behaviour, 24, 119-136. Hanson, R. K., & Wallace-Capretta, S. (2000). Predicting recidivism among male batterers. (User Report 2000-06). Ottawa: Solicitor General Canada. Hare, R. D. (1990). The Hare Psychopathy Checklist - Revised. Toronto: Multi-Health Systems. Harris, G. T., Rice, M. E., & Quinsey, V. I. (1993). Violent Recidivism of mentally disordered offenders: the development of a statistical prediction instrument. Criminal Justice and Behavior, 20, 315-335. Hilton, N. Z., Harris, G. T., Rice, M. E., Lang, C., & Cormier, C. A. (in press). A brief actuarial assessment for the prediction of wife assault recidivism: the ODARA. Psychological Assessment. Kroner, D. G., & Loza, W. (2001). Evidence for the efficacy of self-report in predicting nonviolent and violent criminal recidivism. Journal of Interpersonal Violence, 16, 168-177. Kropp, P. R., & Hart, S. D. (2000). The Spousal Assault Risk Assessment (SARA) guide: Reliability and validity in adult male offenders. Law and Human Behavior, 24, 101-118. Kropp, P. R., Hart, S. D., Webster, C. D., & Eaves, D. (1999). Manual for the Spousal Assault Risk Assessment Guide. Toronto: Multi-Health Systems. Loza, W. & Loza-Fanous, A. (2001). Predictive validity of the self-appraisal questionnaire (SAQ): A tool for assessing violent and nonviolent release failures. Journal of Interpersonal Violence, 15, 1183-1191. Sabourin, D. (1986). Validation of Manitoba probation risk-assessment scale. Winnipeg: Community and Youth Corrections.
|
|