Predicting Juvenile Offender Recidivism: Risk-Need Assessment and Juvenile Justice Officers

Article excerpt

This study compared the validity of structured versus unstructured risk predictions. Fourteen Juvenile Justice Officers (JJOT) provided 121 risk need inventory assessments for offending youth under community supervision. Using a rating scale (0-10), another group of 20 J JOT provided estimates of reoffending for a different sample of 107 young offenders. Follow-up data from a few months to approximately 2 years were accessed to identify new convictions. Predictive validity analyses yielded marginally higher, but non-significant, differences in favour of the inventory (area under ROC curve = .75 versus .70). Detection indices improved slightly when reoffending status was determined for a subset of male offenders at 15 months post-assessment. It was noteworthy that when risk scores were converted to categorical ranges (low, medium, high), the distribution was significantly different for the inventory compared with the officers. This was due to more high-risk ratings by JJOs compared with more low-risk rating from the inventory and highlights one advantage of the latter approach.

Predicting Juvenile Offender Recidivism: Risk--Need Assessment and Juvenile Justice Officers

Assessing the risks and needs of young offenders has become standard practice in many juvenile justice jurisdictions. The Office of Juvenile Justice and Delinquency Prevention in the United States adopted prevention and intervention as the basis for its comprehensive juvenile crime strategy (Coolbaugh & Hansel, 2002; Howell, 1995). A central feature of that strategy involves assessing risk factors and treatment needs which is an approach echoed by many others (Hoge & Andrews, 1996; Day, Howells, & Rickwood, 2004; McLaren, 2000). This widespread juvenile justice strategy reflects some of the same principles (risk, need, responsivity) that have been promoted for 20 years in adult corrections (Andrews, 1991; Andrews & Bonta, 1994) and can be traced to much earlier criminological views (Glueck & Glueck, 1974; Warner, 1923). Assessment inventories for juvenile offenders have been developed to support the risk--need framework in jurisdictions in Canada (Hoge & Andrews, 2003), the United States (Gavazzi et al., 2003; Howell, 1995; Schwalbe, Fraser, Day, & Arnold, 2004), England and Wales (Baker, Jones, Roberts, & Merrington, 2003) and Australia (Thompson & Putnins, 2003).

There are numerous advantages to the systematic assessment of risk and needs using inventories such as those listed above (Hoge, 2002; Thompson & Putnins, 2003). One of the presumed advantages is that future offending is more accurately identified by such inventories and the consequent empirical database than by the idiosyncratic prognostications of juvenile justice staff or clinical experts, for that matter. This argument draws its support from a large body of literature showing that actuarial aggregation of pertinent information matches or betters intuitive judgment when it comes to predicting future behaviours (Bonta, Law, & Hanson, 1998; Dawes, Faust, & Meehl, 1989; Westen & Weinberger, 2004). In spite of the weight of this evidence, predictive validity for any instrument cannot be assumed. Risk-need inventories tend to be developed and adapted in various ways in various jurisdictions and used by staff with differing degrees of training and quality control support (Flores, Travis, & Latessa, 2003).

Studies show that scores produced by risk-need inventories are associated with recidivism for both adult (Waiters, 2006) and juvenile offenders (Jung & Rawana, 1999; National Council on Crime and Delinquency, 2000; Risler, Sutphen, & Shields, 2000; Schwalbe et al., 2004). A meta-analysis (Cotde, Lee, & Heilbrun, 2001) of six predictive validity studies (published between 1988 and 1999) dealing with juvenile recidivism found an overall weighted effect size (Fisher's Z) of. 12 (equivalent to a correlation of .12). However, predictive validity was likely undermined by including studies that used a combination of available variables rather than a formal risk instrument (Katsiyannis & Archwamety, 1997) or studies that concurrently investigated competing predictor variables (Dembo et al., 1998). More impressive predictive validity has been reported in recent studies with correlation coefficients typically around .3 and area under the ROC curve around .70 (Baker et al., 2003; Catchpole & Gretton, 2003; Putnins, 2005; Schmidt, Hoge, & Gomes, 2005; Thompson & Pope, 2005). An exception to these results was reported by Marczyk, Heilbrun, Lander and DeMatteo (2003) who found little evidence of predictive validity in their study of two such inventories. However, their findings may be interpreted as highlighting the threats to validity that can occur with an atypical sample, suboptimal assessment information and retrospective scoring.

Few studies in juvenile justice have compared risk--need inventories with unaided judgements. In an analogue study using videotaped case scenarios of juvenile offenders, court counsellors made unaided risk estimates which were compared to estimates using a structured inventory (Schwalbe et al., 2004). The inventory reduced variation in risk estimates. This supports the claim that a structured approach improves consistency but falls short of confirming improved predictive accuracy. Krysik and LeCroy (2002) compared the predictive accuracy of two risk prediction formulae with estimates by juvenile probation officers. The first actuarial combination was based on 9 offence-related variables; the second used 5 items, 4 of which were dynamic variables. Probation officers judged whether recidivism risk was unlikely (low), chance (medium) or likely (high). Assessments appear to have been made prior to official disposition and officers also completed the 9-item actuarial instrument. Follow-up data (1-2 years) were obtained on over several thousand cases, with recidivism defined as a subsequent complaint. Predictive accuracy was judged somewhat subjectively in terms of the classification rates and associated recidivism rate. In brief, the 5-item statistical prediction performed better than the probation officers who in turn were better than the 9-item prediction. The authors point out that several sources of assessment error undermined accuracy of the 9-item formula.

In Australia, some studies with adult offenders have also addressed structured versus subjective risk assessment. Cumberland and Boyle (1997) examined predictive validity for an adapted version of the Level of Service Inventory (LSI). This was compared with LSI scores finalised by community corrections officers--a proportion of which incorporated subjective adjustments based on additional considerations. Both sets of risk assessments produced equivalent predictive indices for number and severity of reoffences. There was some evidence that adjusted scores added to the predictive discrimination of medium- versus high- risk groups. Unfortunately, the authors provide scant information about the prevalence or extent of score adjustments. Parker (2002) also compared probation/parole officer ratings of recidivism (0-100%) with the Level of Service Inventory--Revised (Andrews & Bonta, 2000). Validity coefficients for a 1-year follow-up period were not significantly different. Parker speculated that staff estimates were well informed as a result of substantial training with the LSI-R and familiarity with the clientele in a small jurisdiction. It might also be added that in the majority of cases, staff had completed the LSI-R with clients for whom they made predictions.

The current study compared the predictive validity of structured versus unstructured risk assessments for juvenile offenders. The study improves on previous investigations by arranging a greater degree of separation between the methods of risk estimation and between the individuals providing them.

Method

Participants

Participants were 34 Juvenile Justice Officers (JJOs) (1) working for the Department of Juvenile Justice, NSW. Of these participants, 14 (9 male, 5 female) were asked to provide assessments using a structured inventory. They were selected to do so because they were familiar with the inventory having participated in an earlier departmental trial. On average, they had previously completed approximately 14 inventories each (range 8 to 28 uses). Staff in this subgroup were from three offices providing community supervision of juvenile offenders in Sydney. Length of experience as a JJO ranged from 3 months to 15 years (M = 7.46, Mdn = 7.50, SD = 5.48). The other 20 JJOs (12 male, 8 female) were asked to provide unstructured risk assessments. They were selected to do so as 11 had not been introduced to the inventoried approach and the remaining nine had used it on average only once or twice (range of 0 to 7 uses). Experience as a JJO ranged from 2 months to 16 years (M = 7.06, Mdn = 6.00, SD = 5.35). Staff in this subgroup were from offices providing community supervision in 3 Sydney and 9 rural NSW locations. There were no significant differences between the two JJO subgroups in terms of length of experience, age and gender distributions, or a Likert rating of job satisfaction.

Measures

Australian Adaptation of the Youth Level of Service/Case Management Inventory (YLS/CMI-AA).

The YLS/CMI-AA (Hoge & Andrews, 1995) is an adapted version of the Youth Level of Service/Case Management Inventory (Hoge & Andrews, 2002). It is intended for use by professionals working in the juvenile justice sector. The inventory was adapted for the Department of Juvenile Justice, New South Wales, before the commercial availability of the parent version. Thompson and Pope (2005) provide details on the adaptation and preliminary psychometric findings from a trial study. The current research used the trial version of the inventory that varied minimally in content from the final adaptation. The trial version was printed in hard-copy format (4 x A4 page booklet). The trial version, like the adapted version and parent inventory, incorporated risk-need items in eight domains: (1) prior and current offences (8 items), (2) family and living circumstances (7 items), (3) education/employment (7 items), (4) peer relations (4 items), (5) substance abuse (6 items), (6) leisure/recreation (3 items), (7) personality/behaviour (7 items), (8) attitudes/orientation (5 items). A new domain, 'Assessment of major strengths', included 3 items related to protective factors. Items are scored in a binary fashion to indicate whether the operationally defined item describes the young person. One item related to age at first court order was scored 0, 1, and 2, with more weight given to younger offenders. Endorsed items in each domain are tallied to provide a domain total and an overall risk-need score is calculated based on all domains except 'Major strengths'.

Reoffending Risk Estimation Scale. This was a Likert scale which called for a rating between 0 and 10 to indicate likelihood of recidivism. Essentially the rating scale was designed to distil subjective considerations of reoffending potential based on knowledge of the client. On the scale, 0 was labelled with the anchor 'No risk of reoffending', and 10 was labelled with the anchor 'Highest risk of reoffending'. Verbal and written information provided to participants emphasised that a subjective estimate was required (e.g., 'Please circle the number which you think is closest to the risk level of your client').

Recidivism. A new criminal conviction subsequent to the date of risk assessment defined recidivism. Breach of an existing order was not counted as reoffending. Length of follow-up was the time between date of risk assessment and date of follow-up or the young person's 18th birthday. Conviction data was only available from the juvenile justice database for the state jurisdiction which with few exceptions was limited to offences committed up to 18 years.

Procedure

The study was approved by the university human ethics committee and by the research ethics committee at the Department of Juvenile Justice, New South Wales. Juvenile Justice Officers volunteered to participate. Risk assessments were completed in 2001 to 2002. Client records were accessed by the Department of Juvenile Justice. In a secure setting, offence information was provided to the authors in hard copy format with client identification numbers.

Results

Predictive Validity

YLS/CMI-AA. Completed inventories and recidivism data were available for 113 young persons (99 male, 14 female) ranging in age from 13.54 to 18.09 years (M = 16.24, SD = 1.08). The number of inventories completed per Juvenile Justice Officer ranged from 4 to 21 (Mdn = 6.5) and the follow-up period varied between 1 and 29 months (M = 16.55, Mdn = 19, SD = 6.97). In this sample, there were 58 recidivists and 55 non-recidivists for a reoffence rate of 51.33%. The mean time to reconviction was 7.34 months (Mdn = 6.0, range 1-20, SD = 4.64). The inventory total score was significantly correlated with recidivism ([r.sub.pb] = .43, p < .001) and area under the receiver operating characteristics curve (AUC; Rice & Harris, 1995) was .75 (p < .001). This indicates that the probability a randomly selected recidivist will have a total score higher than a randomly selected non-recidivist is 75%. All YLS/CMI-AA domain totals except for 'Peer relations' also had significant validity coefficients. The indices of detection performance for 'Prior and current offences' ([r.sub.pb] = .45, AUC = .75, p <.001) were equivalent to the YLS/CMIAA total score. The 'Major strengths' domain was significantly related to recidivism in the expected direction ([r.sub.pb] = -.32; AUC = .33, p < .001). The validity analyses were recalculated for a subset of the follow-up data in order to limit the predictive reach of the inventory to approximately 1 year. Thus, reoffence status was determined at 15 months post-YLS/CMI-AA completion. This timeframe increased the likelihood that offences committed in the first 12 months were finalised and entered on the database. Only reoffences registered between 3 and 15 months were considered because convictions 1 or 2 months post-YLS/CMI-AA were quite likely for outstanding charges prior to completion of the inventory (pseudo convictions). Using these parameters, data were available for 79 young persons (69 males, 10 females) and the analyses were further refined to consider only the subsample of males. The resulting validity coefficients for the YLS/CMI-AA total score are summarised in Table 1 with those from the larger sample for comparative purposes. There appears to be some improvement in detection indices, although the confidence intervals overlap considerably (but these are not independent samples). The pattern of results in the restricted sample for the YLS/CMI-AA domain scores was similar to those in the larger sample with 'Peer relations' being the only non-significant predictor. However, 'Personality and behaviour' was the strongest domain predictor ([r.sub.pb] = .51; AUC = .81, p < .001) and detection indices for 'Prior and current offences' were [r.sub.pb] = .48, AUC = .78, (p < .001).

Risk Estimation Scale. Completed risk estimate ratings and recidivism data were available for 100 young persons (86 male, 14 female) ranging in age from 12.64 to 17.68 years (M= 15.73, SD = 1.10). The number of ratings completed per Juvenile Justice Officer ranged from 1 to 11 (Mdn = 5.0) and the follow-up period varied between 5 and 25 months (M = 17.42, Mdn = 16.5, SD = 4.96). In this sample, there were 53 recidivists and 47 non-recidivists for a reoffence rate of 53%. The mean time to reconviction was 7.43 months (Mdn = 6.0, range = 2-21, SD = 4.74). Risk estimate scores (2) provided by the Juvenile Justice Officers were significantly correlated with recidivism ([r.sub.pb] = .36, p < .001) and AUC was .70 (p = .001). Predictive validity indices for these risk ratings were recalculated for a subset of the follow-up data using the same parameters as were applied to refine the sample for the YLS/CMI-AA. The resulting validity coefficients for 64 males are summarised in Table 1. It can be seen that the detection indices for officer risk ratings are similar in the larger and more restricted samples. The AUC statistic for the officer risk estimates is not significandy different from the corresponding AUC value for the YLS/CMI-AA total score as the confidence intervals overlap.

Categorical Comparison of YLS/CMI-AA and Risk Estimate Scale. Results from the two risk assessment procedures were converted to categorical scores. The possible score range for each risk measure was divided as equally as possible into 3 bands to reflect a low, medium and high score range on each scale. The distributions of these categorical scores are summarised in Table 2 for both the larger and more restricted samples that were previously examined. The per cent of juveniles reoffending during the follow-up period is also shown within each score band in Table 2. Recidivism was significantly associated with risk categorisation in all instances, [chi square] (2, N = 113) = 18.24, p < .001; [chi square] (2, N = 69) = 26.18, p < .001; [chi square] (2, N = 100) = 9.69, p < .01; [chi square] (2, N= 64) = 8.82, p < .05; and the effect size varied from medium to large. The distribution of low-, medium- and high-risk scores was significantly related to the approach to risk assessment (YLS/CMI-AA versus JJO ratings) whether comparing across the two larger samples, [chi square] (2, N= 213) = 14.96,p < .001, or the two more restricted samples, [chi square] (2, N= 133) = 11.46,p < .01. This finding is attributable to the larger proportion of offenders in the low score band based on the YLS/CMI-AA compared with the larger proportion of offenders in the high score band based on global risk estimate ratings. Finally, in each score band, the proportion of recidivists versus non-recidivists did not differ significantly based on the approach to risk estimation.

Discussion

Within the juvenile justice sector it is important to identify young people who are at risk of becoming repeat offenders. It is well recognised that a relatively small proportion of persistent offenders are responsible for a high proportion of crime (McLaren, 2000). There is a longstanding criminological tradition of prediction instruments for this very task (Glueck & Glueck, 1950, 1974) and contemporary versions include assessment of both the risks and needs of juvenile offenders. It is incumbent on those using such inventories to thoroughly analyse their performance characteristic and utility. The current study provides useful information about risk prediction with and without a structured assessment inventory in an Australian juvenile justice setting.

Estimates of risk, based on the structured YLS/CMI-AA or global JJO rating, were equivalent in accurately discriminating between recidivists and non-recidivists. Validity coefficients in this study were similar to those found in other studies (typically correlation coefficients .3 to .4 and area under ROC in the range of .60 to .80) using various risk inventories (e.g., Baker et al., 2003; Catchpole & Gretton, 2003; Flores et al., 2003, Putnins, 2005; Schmidt et al., 2005; Thompson & Pope, 2005). The results in Table 1 suggest that validity coefficients may improve well beyond large effect sizes (Rice & Harris, 2005), particularly for a structured inventory, when forecasts are restricted to a limited timeframe (e.g., 1 year) and re-conviction data allows for typical administrative processing delays. It is also worth noting that a similar degree of predictive accuracy to the Total Score of the YLS/CMI-AA was achieved with the 8 historical items from the 'Prior and current offences' domain. Static risk factors are especially useful in predicting recidivism (Bonta, 2002; Cottle et al., 2001; Gray et al., 2004) and a case can be made for separating risk assessment from needs assessment (Howell, 1995; Thompson, 2005; Thompson & Pope, 2005).

To some it may be surprising that the structured inventory and officer ratings were similar in their predictive utility. It is widely reported that actuarial methods outperform subjective approaches. However, equivalency is not an uncommon outcome in comparison studies (Dawes et al., 1989; Grove & Meehl, 1996; Kleinmuntz, 1990; Westen & Weinberger, 2004). The current study compared informal aggregation of information with structured aggregation of information. The type of information covered by the inventory overlapped considerably with information that was normally collected by Juvenile Justice Officers according to departmental guidelines for background report writing. This included information about offence history and psychosocial functioning and may have contributed to similar predictive success. However, the two approaches to aggregation were not tested for the same individuals on the same information set--a condition that has been stipulated for a fair test of two prediction methods (Dawes et al.). From a methodological point of view, the predictive comparisons of the current study are compromised by using different samples of JJOs and different samples of offenders. The fact that JJOs contributed on average five or six cases to the data sets also means that data points are not independent. Nevertheless, the study had good ecological validity based on real cases in situ and representing distinguishable methods of risk estimation.

It needs to be kept in mind that the risk-need inventory approach rests on the practitioner collecting relevant information (in part via interview) and making judgements about individual inventory items which are then combined quantitatively. This is consistent with what Meehl (1954) referred to as levels of data and Westen and Weinberger (2004) characterise as the actuarial aggregation of clinical data. Such mixed approaches may offer the best of both prediction worlds but did not result in clear superiority of the inventoried approach in the current study. Juvenile Justice Officers participating in this research were aware that prediction methods would be compared and even the group providing unstructured ratings had some exposure to the inventoried approach. This, as well as the prevalent risk-factor discourse within contemporary justice systems, may have contributed to similar predictive outcome. It must also be said that practitioners in the juvenile justice field, unlike many clinicians, have frequent opportunities to see which juveniles reoffend when recidivists return with new charges and supervision requirements. Lack of feedback is one of the characteristics that undermines subjective predictions (Dawes et al., 1989), but may be ameliorated in some circumstances where individuals are 'well-calibrated' for the prediction task (Kleinmuntz, 1990). However, Grove and Meehl (1996) argue that such feedback is unlikely to lead to superior subjective prediction.

Although the risk-need inventory and Juvenile Justice Officers were similar in their overall ability to predict reoffending, the two approaches did differ in an important way. A key finding in the current study relates to the distribution of categorical risk scores that varied with the approach to prediction. The trend was for more offenders in the high-risk category based on the ratings of Juvenile Justice Officers compared with more in the low-risk category based on the inventory. Reoffending rates within these categories were not statistically different. Thus, officers making unaided ratings of risk achieved their overall predictive accuracy using a conservative strategy of concern about reoffending potential. It is true that these categories were imposed post hoc on the data by the authors and not descriptors used by participants in this study. However, score bands and such descriptors are commonly employed with risk-need inventories, albeit based on normative data. A tendency toward conservative bias in clinician judgement (inflated risk estimates) is well recognised in forensic settings (Fuller & Cowan, 1999; Webster, Harris, Rice, Cormier, & Quinsey, 1994). In the current study, the structured inventory achieved equivalent accuracy with 15% to 19% fewer youth in the high-risk category. This is an important difference as there are significant net-widening and resource implications if high-risk cases are to receive the attention they should. For this reason alone, the structured approach to risk classification is to be preferred. The transparency of the inventoried approach which facilitates analysis and evaluation (Thompson, 2006) is further incentive for policy makers in juvenile justice to favour structured risk-need assessment.

Acknowledgment

The authors would like to thank the Collaborative Research Unit and Juvenile Justice Officers of the NSW Department of Juvenile Justice for their assistance in undertaking this research. The opinions here do not necessarily reflect the views of the NSW Department of Juvenile Justice, or any of its officers. References

Andrews, D.A. (1991). Recidivism is predictable and can be influenced: Using risk assessments to reduce recidivism. Forum on Corrections Research, 1(2), 11-17.

Andrews, D.A., & Bonta, J. (1994). The psychology of criminal conduct. Cincinnati, OH: Anderson Publishing.

Andrews, D.A., & Bonta, J.L. (2000). The Level of Service Inventory-Revised. User's manual. North Tonawanda, NY: Multi-Health Systems Inc.

Baker, K., Jones, S., Roberts, C., & Merrington, S. (2003). ASSET-The evaluation of the validity and reliability of the Youth Justice Board's assessment for young offenders. Findings from the first two years of the use of ASSET. Oxford: Centre for Criminological Research.

Bonta, J. (2002). Offender risk assessment: Guidelines for selection and use. Criminal Justice and Behavior, 29, 355-379.

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.

Catchpole, R.E.H., & Gretton, H.M. (2003). The predictive validity of risk assessment with violent young offenders. A l-year examination of criminal outcome. Criminal Justice and Behavior, 30, 688-708.

Coolbaugh, K., & Hansel, C.J. (2002, March). The comprehensive strategy: Lessons learned from the pilot sites. Juvenile Justice Bulletin. Washington, DC: US Office of Juvenile Justice and Delinquency Prevention. Retrieved August 15, 2006, from http://www.ncjrs.gov/html/ojjdp/jjbul2000_03_1/contents.html

Cottle, C.C., Lee, R.J., & Heilbrun, K. (2001). The prediction of criminal recidivism in juveniles: A meta-analysis. Criminal Justice and Behavior, 28, 367-394.

Cumberland, A.K., & Boyle, G.J. (1997). Psychometric prediction of recidivism: Utility of the risk needs inventory. The Australian and New Zealand Journal of Criminology, 30, 72-86.

Dawes, R.M., Faust, D., & Meehl, P.E. (1989). Clinical versus actuarial judgement. Science, 243, 1668-1674.

Day, A., Howells, K., Rickwood, D. (2004). Current trends in the rehabilitation of juvenile offenders. Trends & Issues in Crime and Criminal Justice No. 284. Canberra, Australian Capital Territory, Australia: Australian Institute of Criminology. Retrieved April 4, 2005, from http://www.aic.gov.au/publications/tandi2/tandi284t.html

Dembo, R., Schmeidler, J., Nini-Gough, B., Chin Sue, C., Borden, P., & Manning, D. (1998). Predictors of recidivism to a juvenile assessment center: A three year study. Journal of Child & Adolescent Substance Abuse, 7, 57-77.

Flores, A.W., Travis III, L.F., & Latessa, E.J. (2003). Case classification for juvenile corrections: An assessment of the Youth Level of Service/Case Management Inventory (YLS/CMI). Cincinnati, OH: Centre for Criminal Justice Research, University of Cincinnati.

Fuller, J., & Cowan, J. (1999). Risk assessment in a multi-disciplinary forensic setting: Clinical judgement revisited. The Journal of Forensic Psychiatry, 10, 276-289.

Gavazzi, S.M., Slade, D., Buetmer, C.K., Partridge, C., Yarcheck, C.M., & Andrews, D.W. (2003). Toward conceptual development and empirical measurement of global risk indicators in the lives of court-involved youth. Psychological Reports, 92, 599-615.

Glueck, S., & Glueck, E. (1950). Unraveling juvenile delinquency. Cambridge, MA: Harvard University Press.

Glueck, S., & Glueck, E. (1974). Of delinquency and crime: A panorama of years of search and research. Springfield, IL: Charles C. Thomas.

Gray, N.S., Snowden, R.J., MacCulloch, S., Phillips, H., Taylor, J., & MacCulloch, M.J. (2004). Relative efficacy of criminological, clinical, and personality measures of future risk of offending in mentally disordered offenders: A comparative study of HCR-20, PCL:SV, and ORGS. Journal of Consulting and Clinical Psychology, 72, 523-530.

Grove, W.M., & Meehl, P.E. (1996). Comparative efficiency of informal (subjective, impressionistic) and formal (mechanical, algorithmic) predictions procedures: The clinical-statistical controversy. Psychology, Public Policy, and Law, 2, 293-323.

Hoge, R.D. (2002). Standardized instrument for assessing risk and need in youthful offenders. Criminal Justice and Behavior, 29, 380-396.

Hoge, R.D., & Andrews, D.A. (1995). Australian Adaptation of the Youth Level of Service/Case Management Inventory. North Tonawanda, NY: Multi-Health Systems Inc.

Hoge, R.D., & Andrews, D.A. (1996). Assessing the youthful offender.. Issues and techniques. New York: Plenum.

Hoge, R.D., & Andrews, D.A. (2003). Youth Level of Service/Case Management Inventory. User's manual. North Tonawanda, NY: Multi-Health Systems.

Howell, J.C. (1995). Guide for implementing the comprehensive strategy for serious, violent, and chronic juvenile offenders. Washington, DC: Office of Juvenile Justice and Delinquency Prevention.

Jung, S., & Rawana, E.P. (1999). Risk and need assessment of juvenile offenders. Criminal Justice and Behavior, 26, 69-89.

Katsiyannis, A., & Archwamety, T. (1997). Factors related to recidivism among delinquent youths in a state correctional facility. Journal of Child and Family Studies, 6, 43-55.

Kleinmuntz, B. (1990). Why we still use our heads instead of formulas: Toward and integrative approach. Psychological Bulletin, 107, 296-310.

Krysik, J., & LeCroy, C.W. (2002). The empirical validation of an instrument to predict risk of recidivism among juvenile offenders. Research on Social Work Practice, 12, 71-81.

Marczyk, G.R., Heilbrun, K., Lander, T., & DeMatteo, D. (2003). Predicting juvenile recidivism with the PCL:YV, MAYSI, and YLS/CMI. International Journal of Forensic Mental Health, 2, 7-18

McLaren, K.L. (2000, June). Tough is not enough--Getting smart about youth crime. A review of research on what works to reduce offending by young people. Wellington, New Zealand: Ministry of Youth Affairs.

Meehl, P.E. (1954). Clinical versus statistical prediction. A theoretical analysis and a review of the evidence. Minneapolis, MN: University of Minnesota Press.

National Council on Crime and Delinquency (2000). Almeda county placement risk assessment validation. Draft of Final Report. Oakland, CA: Author.

Parker, R.J. (2002). Intersource agreement on the prediction of recidivism. Unpublished master's thesis, University of Canberra, Canberra, Australian Capital Territory, Australia

Putnins, A.L. (2005). Assessing recidivism risk among young offenders. The Australian and New Zealand Journal of Criminology, 38, 324-339.

Rice, M. E., & Harris, G.T. (1995). Violent recidivism: Assessing predictive validity. Journal of Consulting and Clinical Psychology, 63, 737-748.

Rice, M.E., & Harris, G.T. (2005). Comparing effect sizes in follow-up studies: ROC area, Cohen's d, and r. Law and Human Behavior, 29, 615-620.

Risler, E.A., Sutphen, R., & Shields, J. (2000). Preliminary validation of the juvenile first offender risk assessment index. Research on Social Work Practice, 10, 111-126.

Schmidt, F., Hoge, R.D., & Gomes, L. (2005). Reliability and validity analyses of the Youth Level of Service/Case Management Inventory. Criminal Justice and Behavior, 32, 329-344.

Schwalbe, C.S., Fraser, M.W., Day, S.H., & Arnold, E.M. (2004). North Carolina Assessment of Risk (NCAR): Reliability and predictive validity with juvenile offenders. Journal of Offender Rehabilitation, 40, 1-22.

Thompson, A.P. (2005). Risk-need assessment with juvenile offenders: Past, present and future. In M. Katsikitis (Ed.), Proceedings of the 40th APS Annual conference (pp. 343-347). Melbourne, Victoria, Australia: The Australian Psychological Society Ltd.

Thompson, A.P. (2006). Do we need to worry about subgroup differences in juvenile offender risk-need assessment? In M. Katsikitis (Ed.), Proceedings of the 2006 Joint Conference of the APS and NZPsS (pp. 445-449). Auckland, New Zealand: The Australian Psychological Society Ltd.

Thompson, A.P., & Pope, Z. (2005). Assessing juvenile offenders: Preliminary data for the Australian Adaptation of the Youth Level of Service/Case Management Inventory (Hoge & Andrews, 1995). Australian Psychologist, 40, 207-214.

Thompson, A.P., & Putnins, A.L. (2003). Risk-need assessment inventories for juvenile offenders in Australia. Psychiatry, Psychology and Law, 10, 324-333.

Upperton, R., & Thompson, A.P., (2003). Risk assessment in juvenile justice: Predictive and practical utility. In M. Katsikitis (Ed.), Proceedings of the 38th APS Annual Conference (pp. 236-241). Melbourne, Victoria, Australia: Australian Psychological Society Ltd.

Warner, S.B. (1923). Factors determining parole from the Massachusetts reformatory Journal of Criminal Law & Criminology, 14, 172-207.

Waiters, G.D. (2006). Risk-appraisal versus self-report in the prediction of criminal justice outcomes. A meta-analysis. Criminal Justice and Behaviour, 33, 279-304.

Webster, C.D., Harris, G.T., Rice, M.E., Cormier, C., & Quinsey, V.L. (1994). The violence prediction scheme: Assessing dangerousness in high risk men. Toronto, Canada: Centre of Criminology, University of Toronto.

Westen, D., & Weinberger, J. (2004). When clinical description becomes statistical prediction. American Psychologist, 59, 595-613.

Rachel A. Upperton and Anthony P. Thompson

Charles Sturt University, Australia

Correspondence to: Dr Anthony P. Thompson, Humanities and Social Sciences, Charles Sturt University, Locked Bag 678, Wagga Wagga NSW 2678, Australia E-mail: athompson@csu.edu.au

Endnotes

(1) These Juvenile Justice Officers were volunteers from a group of 40 who had participated in a related qualitative study by the authors. They had previously been interviewed about how they normally determine whether a young person is going to reoffend and their views about using a structured risk inventory. Preliminary results were reported by Upperton and Thompson (2003).

(2) Scores were coded 0-9 because the scale was inadvertently printed with 9 rather than 10 equal intervals.

Table 1
Comparative Predictive Validity of Risk Assessments

Predictor          Follow-up       Time to            Sample
                     period      reconviction
                    (months)       (months)

YLS/CMI-AA (a)        1-29           1-20       99 male, 14 female
YLS/CMI-AA (a)         15            3-15            69 male
Officer Rating        5-25           2-21       86 male, 14 female
Officer Rating         15            3-15            64 male

Predictor         Reconviction   [r.sub.pb]     AUC      SE    95% CI
                      rate

YLS/CMI-AA (a)        51%          .43 **     .75 ***   .05   .66-.84
YLS/CMI-AA (a)        57%          .57 **     .83 ***   .05   .73-.92
Officer Rating        53%          .36 **     .70 ***   .05   .60-.80
Officer Rating        61%          .42 **     .73 **    .06   .61-.86

Note: AUC = area under ROC curve: CI = confidence interval

(a) Total score

** p < .01 *** p [less than or equal to] .001

Table 2
Comparative Distribution of Scores and Recidivists for
Risk Assessment

Measure              N         Score range

                            Low   Medium   High

YLS/CMI-AA (a)      113     36%      43%    21%
Recidivism           58     27%      58%    79%
YLS/CMI-AA (a)       69     33%      46%    20%
Recidivism           39     17%      66%   100%

Officer rating      100     14%      50%     36
Recidivism           53     21%      50%    69%
Officer rating       64     11%      50%    39%
Recidivism           39     14%      59%    76%

Note: Row data coincide with the follow-up, reconviction and sample
characteristics in Table 1. Recidivism per cent is within each score
range

(a) Total score.