Using NIBRS Data to Model the Predictive Ability of Case Based Reasoning for Repeat Victimizations

Cynthia Blackburn Line, Rowan University
Michael Redmond, La Salle University

ABSTRACT
Research, while limited, continues to demonstrate that repeat victimizations appear to be unique compared to other crimes. This uniqueness obliges researchers to look past traditional approaches (e.g. demographic or geographic variables). While current statistical models explain variation in dependent variables, they cannot predict retroactively or proactively, nor can they test the accuracy of predictions. The accuracy of current statistical techniques is further limited by various problems, including underlying assumptions. Case-Based reasoning (CBR), a development in the use of artificial intelligence, allows for better predictive analyses and is not plagued by statistical problems. CBR, used on victimization data, may predict whether an individual would be a multiple victim. Previous research utilizing data from the NCVS demonstrated some success with this approach. However, the NCVS data themselves are problematic for various reasons. Problems with the NCVS data stem primarily from a lack of incident based data which might be necessary for appropriate model formation. This project utilizes CBR to analyze victimization information from the NIBRS data. The NIBRS data are a more robust data set, featuring many different variables previously unavailable in victimization studies using the NCVS data. In addition, this project includes additional learning features designed to improve the accuracy of predictions. A successful series of retroactive predictions will allow for greater learning and refined models. If the program can successfully predict multiple victimizations retroactively, there exists potential for proactive prediction with proper model specification.

(Return to Program Resources)

Updated 05/20/2006