Modeling the Referral Decision in Sexual Assault Cases: An Application of Random Forests

Abstract

This paper examines the decision to refer a sexual assault case for prosecution using a sample of 730 reported sexual assaults in which the victim received a medical/forensic examination. The decision to refer a case for prosecution was modeled using an algorithmic modeling technique, Random Forests. The key advantages of this modeling approach include its superiority in predicting case outcomes and its ability to easily uncover nonlinear relationships. Key results indicate that the likelihood of referral increased when sperm was found and documented, when the victim could identify the suspect, and as the severity of nongenital injury increased. Neither the presence nor the severity of genital injury impacted the decision to refer a case for prosecution. On the whole, suspect and report characteristics had the largest impact on referring cases for prosecution, with victim characteristics having little influence.