Вы находитесь на странице: 1из 4

Article Summary - Thirty Years of Research on the Level of Service Scales: A Meta-Analytic Examination of Predictive Accuracy and Sources

of Variability - Mark E. Olver, Keira C. Stockdale, and J. Stephen Wormith (2013) Level of Service Scales (LS) The LS scales are the most used risk assessment tools (1,085,647 declared administrations in 2010). Several revisions have been published since the original LS Inventory. Of the various versions the LSI-Revised and its short form (LSI-R: Screening Version) are currently the most widely used. -The LS scales follow a risk-need-responsivity (RNR) framework and are based on the Big Four covariates of criminal conduct: criminal history, antisocial attitudes, antisocial associates, and antisocial personality pattern), in addition to education/employment, family/ marital, leisure recreation, and substance abuse. Combined they are referred to the Central Eight. -The LS scales link assessment with intervention. The scales are designed to appraise recidivism risk, identify criminogenic needs (i.e., dynamic risk factors) for intervention, and inform recommendations for treatment, case management, and community supervision. -Big Four and Central Eight underpin a general personality and cognitive social learning theory of criminal behavior that provides an explanatory model of the origin and continuation of criminal conduct, and informs methods for predicting, reducing, managing, and preventing criminal behavior. Predictive Accuracy Though not the only important quality, strong predictive accuracy is a prerequisite for a tools usefulness. Evaluations of the LS scales predictive accuracy have shown them to have high predictive accuracy for general recidivism, and moderate accuracy for prediction of violence. Though some debate exists, in comparison to other tools, the LS scales have generally shown to be at least as accurate as other tools. Applicability for Female Offenders The gender informed position argues that female offenders differ from male offenders in important ways and thus have unique needs. The LS has been criticized for not adequately accounting for these differences and gender-specific responsive needs of female offenders such as parental stress, low self-esteem, childhood, domestic and sexual abuse, anger concerns, and poverty, among others. However, supplemental scales developed to address these areas have shown equivocal results in regard to improving predictive accuracy beyond the regular LS scales. - Countering the gender informed position is the gender neutral position which argues that although important responsivity considerations are required for females, the covariates of criminal behavior are generally the same and thus the RNR model remains applicable. Supporting this position is research demonstrating similarly strong predictive accuracy of the LS scales for both male and female offenders. Despite this support, limited research has been conducted assessing validity for female offenders of the individual need domains of the LS. Applicability for Ethnic Minorities There are concerns about the use of the LS scales and similar tools with ethnic minorities due to a lack of research examining their validity with these

populations. It has been demonstrated in Aboriginal offenders and argued that risk-factors commonly present in some minority groups may artificially inflate risk assessment scores. Other research has shown that predictive accuracy of the LS for general recidivism in Aboriginal offenders is slightly lower, but still significant, and there is an even smaller discrepancy for violent recidivism. Nevertheless, there remains a need for larger-scale quantitative reviews examining LS use with ethnic minority groups. Large-Scale Meta-Analysis -A large-scale meta-analysis was conducted to extend findings on the predictive accuracy of the LS and further address questions surrounding the validity of its use for female and ethnic minority offenders. It also sought to compare the accuracy of the various versions of the LS scales. Methods - 128 studies met the inclusion criteria for the meta-analysis, representing 151 independent samples and 137,931 offenders. Most studies were from Canada (55) and the United States (53), followed by Australia (8), United Kingdom (6), Singapore (2), Germany, Japan, New Zealand, and Pakistan (1 each). 80.5% of the sample was male and 19.5% female. The mean age (unweighted) across the samples was 26.67 years. 63% of total participants were White, 18.9% Black, 9.8% Aboriginal, 5.5% Hispanic, 2.9% Asian, and 6.5% of other ethnic descent Seventytwo percent of the samples featured adult populations, whereas 28% were youth. The mean rates of recidivism were 36% for general, 35.2% nonviolent, 13.7% violent, and 6.5% sexual recidivism. - Predictive accuracy statistics were coded in terms of r, which in most cases was a point biserial correlation, or rpb (i.e., a correlation between a continuous predictor, such as the score on a risk measure, and a binary criterion variable, such as dichotomous recidivism coded yesno). When r was not reported, the appropriate formula was applied to convert the reported statistic or descriptive information (e.g., mean group differences between recidivists and non-recidivists) into r. Both a fixed-effects (larger sample sizes weighted greater) model and a random-effects model (more even weighting across sample sizes). Results and Discussion Gender Males scored slightly higher LS total scores, had a more serious offense history, and more pervasive patterns of antisocial behavior, as well as marginally higher scores in areas related to concerns with antisocial peers, lack of prosocial leisure activities, and substance abuse linked to crime. Females had more serious personal/emotional concerns, financial problems, and family/marital difficulties, and significant but smaller differences indicating greater accommodation and education/employment concerns. The authors deemed the findings to be consistent with the argument for salient areas of risk and need for female offenders, given their unique circumstances and possible gendered pathways to crime. The authors noted that the finding that women scored lower on the substance abuse domain was surprising given the literature indicating that substance abuse plays a critical role in the pathway to female criminality.

In terms of predictive accuracy, the LS total score accuracy was very similar for males and females for predicting general recidivism. The LS domains each significantly predicted violent and general recidivism among both genders, with few differences in effect size magnitude aside from substance abuse and personal/emotional, which had significantly larger effect size for females in the prediction of general recidivism. The authors stated that these findings are in line with the literature and that women as a whole may abuse substances less, but when they do it is particularly problematic. They concluded that overall there is little evidence to suggest that the instrument, overall, is better suited for, or performs better for, either gender group, at least in terms of recidivism prediction. Ethnicity - Ethnic minorities scored higher than non-minorities on LS total score and all but one criminogenic need domain (personal/emotional). The magnitude of the differences was small except for the antisocial pattern domain which was approx. medium size. These results are generally in-line with previous findings, but it is important to consider important social, historical, and contextual factors that potentially contribute to risk score elevation and increase the possibility of contact with the justice system. -In terms of predictive accuracy, the LS and its need domains significantly predicted general and violent recidivism among ethnic minority and non-ethnic minority groups. Notably, lower predictive accuracy of LS total scores was observed with the ethnic minority samples when using the fixed-effects model, but the difference was reduced in the random-effects model, particularly for prediction of violent recidivism. The authors concluded that the results are consistent with past research findings supporting the validity of the Central 8 domains in international samples of ethnic minorities. Findings showing greater accuracy for nonminorities compared to aboriginals were replicated for general recidivism, with little difference between groups for violent recidivism. Effect size difference between nonminority and minority groups decreased when LS total score predictive accuracy and general recidivism was examined. This was most evident for Canadian samples, but heterogeneity continued to exist in U.S. minority and nonminority samples for the prediction of general recidivism. The authors posited that systemic bias may impact measurement of true recidivism for minority offenders in the U.S. Geographical Considerations - The largest effect sizes were observed in Canadian samples, followed by non-North American countries, with U.S. samples demonstrating the lowest effect sizes. Nevertheless, the LS scales and their riskneed domains still significantly predicted all recidivism outcomes irrespective of geographic region. The results point to the importance of geographic region as an important source of variation. The authors posited that differences by region may be due to error in both the nature of the true relationship and the precision of both the assessment and outcome measures could vary by region. They also suggest the influence of cultural differences may be important, citing the Canadian heritage of the LS scales, but do not believe this to be a major concern given the universality of the Central Eight domains. Interpretation Considerations and Limitations The authors claim this to be the largest LS study to date. The authors warn that difference on total risk score of the need domains between males and females or ethnic groups, does not necessarily indicate bias in the instrument. They

suggest it is important to determine whether these differences in risk correspond to differences in outcome in that higher mean risk scores should correspond with higher recidivism rates. They point to findings that ethnic minorities have higher recidivism rates than White offenders, and so on average, higher risk scores may in fact be accurate. -The authors identify limitations of the study as including potential exclusion of foreign language studies and other gray literature. They also note that relatively few investigations reported their results combining the domains of gender and ethnicity instead reporting effect sizes as a function of one broad group, failing to report findings based on ethnicity. They highlight potential confounding moderators including regional findings being impacted by assessor training or caseload size, LS experience, agency quality assurance mechanisms, author affiliation, and precision of the outcome measure. Lastly, the authors admit that they did not attempt to explore the application of LS total and need scores to offender case management and treatment in accordance with third-and fourth-generation risk assessment principles.

Вам также может понравиться