The utility of latent variable models in refining behavioural crime scene analysis of serial stranger sexual offences

Behavioural crime scene analysis (BCSA) is a police tool used to reconstruct an offence based on behaviours. Recently, BCSA has demonstrated clinical utility by predicting recidivism and aiding case conceptualization. However, a systematic review of BCSA models showed a paucity of research evaluating which behaviours are necessary and sufficient to model sexual offences. Groth and Birnbaum’s sex offender typology, which is based on offence behaviours, provides a theoretical framework that integrates investigative information and clinical practice. The purpose of this thesis was to evaluate statistical- and theory-based approaches to refine BCSA models that distinguish sexual offenders. In Studies 1 through 3, Multidimensional Scaling, Nonlinear Principal Component Analysis, and Latent Class Analysis were used to create statistically-driven and theory-driven behavioural models from 59 serial, stranger sexual offenders. Validity testing of the theory-driven model indicated that applying Groth and Birnbaum’s framework to BCSA could optimize both investigative efforts and clinical decision-making.