For Immediate Release: December 24, 2009
Contact: Carla Newman – 303.216.9455
cnewman@npipm.com  

Northpointe's  Research and Development Department presented two papers at the 2009 Annual Meeting of the American Society of Criminology in Philadelphia  

In our paper, “Female Pathways to Prison: From Quantitative Taxonomy to Explanatory Theory and Process Narratives,” we describe the development and validation of eight type-patterns that appear to underlie the heterogeneity of the female prison inmates. This taxonomy is based on multiple quantitative pattern analysis of a sample of 718 female prisoners. The taxonomic profiles address a comprehensive assessment of gender-sensitive factors (parenting issues, sexual abuse, trauma, battering relationships, etc.) as well as a broad set of gender-neutral criminogenic factors (poverty, anti-social peers, educational deficits, etc.). Following a review of validation evidence we explore the links and potential replications of this taxonomy to prior studies of women’s pathways to crime e.g. Daly’s Feminist Pathways, Moffitt’s taxonomy and other recent typologies of female offenders in the recent literature. Evidence of replication of several types in this literature is identified, including analogues of types from both Moffitt’s taxonomy and Daly’s Feminist Pathways. Each female type offers a coherent narrative with specific pathways and patterned events leading to their offence patterns. We conclude by attempting to clarify how each prototypical category of women offenders relates to current criminological theories of women’s offending.

In our second paper, “Prognostic Modeling of Relatively Rare Recidivism Events: Beyond the AUC,” we show several graphical methods for evaluating the accuracy and performance of risk scales. Typically risk scales are evaluated using the receiver operating characteristic (ROC) curve; however, most evaluations are limited to reporting the area under curve (AUC). Although the AUC is a good summary of the accuracy of a risk scale over the range of possible cut-off scores, it does not completely describe the performance of a risk scale. It is well-known that performance also depends on the base rate and the cut-off score used. When predicting relatively rare events, such as violent felony recidivism, it may be difficult for a risk scale to out-perform the naïve prediction (no one fails), even if the scale has good accuracy as measured by the AUC. We use data from a large outcomes study in probation to illustrate the challenges of predicting relatively rare events and demonstrate the use of alternative verification methods for prognostic models in this context.