A Multivariate Model for Analyzing Crime Scene Information

Abstract

This study examines the validity of a statistical offender profiling technique that predicts the multi-dimensional classification of homicide offenders. Analyzing 539 Japanese homicide cases, we constructed multivariate prediction models that infer classifications defined by three dichotomous variables (stranger offender, solo offender, money-oriented motive) on the basis of crime scene information. We evaluated the validity of the models with a 10-fold cross-validation procedure and a receiver operating characteristic (ROC) analysis, and found the models to have moderate accuracy (area under the curve [AUC] = .73 to .82). We discussed the results from the perspective of the offender’s rational choices in the crime scene and crime specialization.