The criminogenic needs of female offenders: Relationships between criminal risk factors and recidivism

Much of the current research on the assessment and treatment of justice-involved individuals includes an emphasis on select risk factors that have a direct relationship to criminal behavior (Andrews, Bonta, & Wormith, 2006), and this research informed the development of the Level of Service Inventory-Revised (LSI-R; Andrews, Bonta, & Wormith, 2004), the most widely used assessment of criminal risk across the globe (Wormith, 2011). The LSI-R, and its subsequent revisions, provide an estimate of recidivism risk based upon the presence or absence of these risk factors. Feminist scholars criticize the use of the Level of Service Inventory family of assessments (LS) for measuring risk in justice-involved females, as it was developed and normed using justice-involved males. Of concern is that the LSI-R is routinely implemented with justice-involved females without proper evidence of its reliability and validity for this unique population (Blanchette & Brown, 2006; Davidson & Chesney-Lind, 2009; Holtfreter & Morash, 2003; Van Voorhis 2012). Although there appears to be strong evidence for the consistency of the predictive ability of the LSI-R across gender, this does not indicate that the instrument measures criminal risk in females the same way as it does for males (Andrews et al., 2012; Holtfreter & Morash, 2003). The purpose of this study was to conduct an item-level examination of the Level of Service Inventory-Ontario Revision (LSI-OR) to assess for gender bias. It was hypothesized that items related to criminal history, antisocial personality pattern, and antisocial thinking would not evidence clinically significant measurement bias. It was also hypothesized that items related to criminal associates would be more relevant for males, whereas items related to substance abuse would be more relevant for females. An item response theory approach was used to identify items with potential bias, and archival data was used to ensure adequate statistical power. The sample consisted of a total of 48,572 individuals with a gender split of 81.6% males (n = 39,640) and 18.4% females (n = 8,932). Results of a series of 2-parameter logistic regression models used to detect the presence of differential item functioning indicated no significant gender differences on any of the items examined. The results of the present study provide additional support for the gender-neutral theory of criminal behavior generally, and the continued use of the LSI-OR in women, specifically. Whether a gender-neutral approach to item construction performs similarly to a gender-responsive approach to item construction (i.e., items on risk assessments designed especially for women) remains to be examined.